Internal Structure and Volcanic Hazard Potential of Mt Tongariro, New Zealand, from 3D Gravity and Magnetic Models

Journal of Volcanology and Geothermal Research 319 (2016) 12–28

Contents lists available at ScienceDirect

Journal of Volcanology and Geothermal Research

journal homepage: www.elsevier.com/locate/jvolgeores

Internal structure and volcanic hazard potential of Mt Tongariro, New Zealand, from 3D gravity and magnetic models

Craig A. Miller a,b,⁎, Glyn Williams-Jones b a Department of Earth Sciences, Simon Fraser University, Burnaby, BC V5A 1S6, Canada b GNS Science, Wairakei Research Centre, Private Bag 2000, Taupo 3352, New Zealand article info abstract

Article history: A new 3D geophysical model of the Mt Tongariro Volcanic Massif (TgVM), New Zealand, provides a high resolu- Received 14 November 2015 tion view of the volcano's internal structure and hydrothermal system, from which we derive implications for Received in revised form 14 March 2016 volcanic hazards. Geologically constrained 3D inversions of potential field data provides a greater level of insight Accepted 16 March 2016 into the volcanic structure than is possible from unconstrained models. A complex region of gravity highs and Available online 19 March 2016 lows (±6 mGal) is set within a broader, ~20 mGal gravity low. A magnetic high (1300 nT) is associated with Keywords: Mt Ngauruhoe, while a substantial, thick, demagnetised area occurs to the north, coincident with a gravity low Gravity and interpreted as representing the hydrothermal system. The hydrothermal system is constrained to the west Magnetic by major faults, interpreted as an impermeable barrier to fluid migration and extends to basement depth. 3D modelling These faults are considered low probability areas for future eruption sites, as there is little to indicate they Volcanic hazard have acted as magmatic pathways. Where the hydrothermal system coincides with steep topographic slopes, Hydrothermal system an increased likelihood of landslides is present and the newly delineated hydrothermal system maps the area Volcanic structure most likely to have phreatic eruptions. Such eruptions, while small on a global scale, are important hazards at the TgVM as it is a popular hiking area with hundreds of visitors per day in close proximity to eruption sites. The model shows that the volume of volcanic material erupted over the lifespan of the TgVM is five to six times greater than previous estimates, suggesting a higher rate of magma supply, in line with global rates of andesite production. We suggest that our model of physical property distribution can be used to provide constraints for other models of dynamic geophysical processes occurring at the TgVM. © 2016 Elsevier B.V. All rights reserved.

1. Introduction with magma pathways provides important information on the likelihood of such eruptions and allows suitable hazard mitigation to be Knowledge of a volcano's internal structure is important for many put in place (Potter et al., 2014). Long-lived hydrothermal systems con- aspects of volcanology and volcanic hazard assessment. This is especial- siderably alter and mechanically weaken large volumes of rock, which if ly so in complex multi-vent systems where there is no central vent coincident with steep slopes presents a considerable landslide, lahar through which most eruptions occur and where multiple vents have and ﬂank collapse hazard (e.g., López and Williams, 1993, Day, 1996; been active in historic times. By geophysically imaging the volcano Finn et al., 2001; Reid et al., 2002; Moon et al., 2005; Tontini et al., 2013). plumbing system and structures in the basement below the volcanic Geophysical knowledge of a volcano's internal physical property ediﬁce, it is possible to assess the importance of these structures in con- distribution also provides context within which processes that occur trolling magma ascent paths and vent locations. In addition, knowledge during volcanic unrest can be interpreted. Often, geophysical models of the extent of a volcano's hydrothermal system provides important in- of volcano unrest are limited by use of an unrealistic uniform halfspace formation on the likely style of eruptions. Hydrothermal systems often or simple 1D model: the necessary geophysical context required for manifest as scenic surface features, attracting hikers and tourists, but more detailed modelling is unknown (Cannavò et al., 2015). This results when over-pressurised can produce small, but dangerous phreatic in inaccurate models which impedes scientists' ability to make informed eruptions with very little warning (e.g., Raoul Island, Christenson decisions during times of volcanic unrest. et al., 2007; Te Maari, Procter et al., 2014;Ontake,Sano et al., 2015) Here we present a new, detailed, 3D geophysical model of the multi- and are often overlooked in volcanic hazard assessments. As such, vent Mt Tongariro volcanic massif (TgVM), New Zealand, combining an knowledge of the extent of a hydrothermal system and its interaction extensive new gravity dataset with aeromagnetic and geological data. We use a geologically constrained inverse modelling technique not previously applied to complex multi-vent andesite stratovolcanoes (cf. ⁎ Corresponding author at: Department of Earth Sciences, Simon Fraser University, Burnaby, BC V5A 1S6, Canada. Blaikie et al., 2014), to produce a geologically sound and geophysically E-mail address: [email protected] (C.A. Miller). accurate model of the TgVM. This model enables examination of

http://dx.doi.org/10.1016/j.jvolgeores.2016.03.012 0377-0273/© 2016 Elsevier B.V. All rights reserved. C.A. Miller, G. Williams-Jones / Journal of Volcanology and Geothermal Research 319 (2016) 12–28 13

1) the basement surface and faulting under the edifice, 2) the bulk inter- east, these are the National Park fault, Waihi fault zone, Poutu fault nal structure of the volcano and 3) the extent of the hydrothermal sys- zone and the inferred location of the northwest dipping Rangipo fault tem. Furthermore, we assess the volcanic hazard implications of (Fig. 1). features in our model. For example the distribution of hydrothermally Jurassic age basement rocks of the Torlesse Terrane outcrop in the altered rock has an influence on future landslide potential and we Kaimanawa Ranges on the east, while Waipapa Terrane rocks outcrop consider the likelihood of basement faults acting as future magma in the far west of the model area. In the centre of the Mt Ruapehu pathways. graben, basement rocks are inferred to be overlain by a thin layer (100 m) of Tertiary sediments. Tunnels drilled as part of the Tongariro 2. Geologic setting and existing geophysical data power scheme in the far north-west of the study area intersected Waipapa Terrane greywacke beneath surface Tertiary sediments at a Interest in the TgVM has increased since early 2000 when unusual depth of around 100 m (Beetham and Watters, 1985). tornillo-type earthquakes were detected (Hagerty and Benites, 2003) Here we refer to the TgVM as the various eruptive centres that make around the Te Maari craters (Fig. 1). In 2005–2009, a long sequence of up Mt Tongariro, including Mt Ngauruhoe (2280 m). The TgVM is con- small volcanic earthquakes occurred close to Mt Ngauruhoe (Jolly structed of at least 17 overlapping vents built during 6 main cone build- et al., 2012), 30 years after its last eruption, and in 2012 two eruptions ing episodes and covers an area of 5 by 13 km (Hobden et al., 1999). The occurred from the Upper Te Maari Crater, the first confirmed eruptions massif has been extensively modified by glaciation since the first erup- from this vent in over 100 years (Scott and Potter, 2014). tions around 275 ka, thus surface exposures of early vents are obscured The TgVM lies at the southern end of the Taupo Volcanic Zone (TVZ), by later eruptions or have been removed by erosion. in a back-arc setting resulting from the westward subduction of the Earliest activity began in the area of Lower Tama Lake (Tama Pacific plate beneath the North Island of New Zealand. Within the 1) (Fig. 1), followed by activity around 200 ka at a nearby centre, back-arc setting is an extensional environment known as the Taupo Tama 2. A long lived cone north of Oturere Valley (Northeastern or Ruaumoko Rift (Rowland and Sibson, 2001, Acocella et al., 2003). Oturere, Mangahouhounui lavas) was built between 105 and 130 ka Extension across this rift is accommodated by segments or domains during which time a vent near Pukekaikiore was also active. Another of sub-parallel north-west and south-east dipping normal faults centre, Tongariro Trig formed between 65 and 110 ka, while contempo- (Seebeck et al., 2014). The TgVM is located at the northern end of the raneously a cone formed to the south of Oturere Valley (Southwestern Ruapehu or Tongariro domain, an area dominated by andesitic volca- Oturere, Waihohonu lavas) (Hobden et al., 1996). Around 25 ka, activity nism, south of the dominantly rhyolitic Taupo domain. The geologic ex- started at Te Maari, Tama Lakes, Red Crater, North Crater, Blue Lake and tension rate across the graben (Mt Ruapehu graben) formed by normal Pukekaikiore (Nairn, 2000). Since around 7 ka, activity has been domi- faulting in the Tongariro domain is estimated by Villamor and nated by the growth of Mt Ngauruhoe cone (Moebis et al., 2011) Berryman, (2006b) to be 2.3 ± 1.2 mm/year. Several sub-parallel faults while historic activity has been from Mt Ngauruhoe, Red Crater and and fault zones delineate the graben in our study area; from west to Upper Te Maari (Scott and Potter, 2014).

Fig. 1. Simplified geological map of the TgVM. The inset map shows the North Island of New Zealand with the TVZ and model area outlined in black. The dashed red line is the Pacific/ Australian plate boundary. Coordinates are easting and northing in m using the NZTM projection. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.) 14 C.A. Miller, G. Williams-Jones / Journal of Volcanology and Geothermal Research 319 (2016) 12–28

Flank collapse has punctuated cone building episodes, either trig- grouped samples into four main rock types: Andesite lava, referring to gered by eruptions (Lecointre et al., 2002), or triggering eruptions by dense lava flows; Pyroclastic, a range of material from pumice and scoria rapidly de-pressurising the hydrothermal system (Jolly et al., 2014). In to denser welded agglutinates; Greywacke, referring to basement both cases, the active hydrothermal system played an important role Torlesse and Waipapa Terrane rocks; and Sandstone, Tertiary sand- in mechanically weakening the rock prior to failure (Breard et al., stones from the Taumarunui Formation. 2014). Currently the largest surface hydrothermal features are at Red For each rock type, we computed physical property histograms Crater, Ketetahi and Te Maari craters, although other mapped areas of (Supplementary Material Fig. 3) with the mean and standard deviation alteration suggest a long history of hydrothermal activity in many loca- for each (Table 1). Wet densities better represent whole rock densities tions on the massif. What is not documented from surface mapping is for rocks that are below the water table and are more suitable for gravity how extensive alteration is within the massif. modelling. Depending on the porosity of the rock, dry vs wet densities in these samples can vary by as much as 340 kg/m3. 2.1. Previous geophysical studies Magnetic susceptibilities of fresh volcanic rock samples range from 0.001 to 0.04 SI, while basement greywacke rocks are only very weakly Previous geophysical studies have imaged the structure, magmatic magnetic (b0.001 SI) or non-magnetic (below detection limit). Hunt and hydrothermal systems of the TgVM at varying degrees of resolution. and Mumme (1986) showed magnetisation intensities of young Mt Zeng and Ingham (1993) undertook two dimensional modelling of Ngauruhoe lavas range from 0.7 to 49 A/m from unweathered samples. sparse gravity data along a profile south of Tama Lakes and suggested No samples of hydrothermally altered rock were available in the data- the presence of low density pyroclastic material overlying a dense base- base and measurements of magnetic susceptibility in volcanic rocks ment. Walsh et al. (1998) summarised electrical resistivity data to delin- may be dominated by remnant magnetisation, so are only used as a eate the extent of the hydrothermal system along a single profile; they guide. found a shallow low resistivity layer, interpreted as geothermal condensate several hundred metres thick, overlying a vapour-dominated layer 3. Geophysical data acquisition and processing of unknown thickness. Rowlands et al. (2005) undertook a moderate resolution seismic tomography study and identified significant low ve- Our study covers 504 km2 within a rectangle 28 km × 18 km ranging locity anomalies beneath Mts Ruapehu, Tongariro and Ngauruhoe in elevation from 600 m to 2300 m. This region encompasses all the lava which they interpreted as remnant magma batches and thick pyroclas- flows from the TgVM and includes basement rocks outcropping to the tic material from various volcanic sources. Cassidy et al. (2009) pro- east and west of the volcano. duced a more detailed 2D model across the Tama Lakes profile, from new gravity, aeromagnetic and magnetotelluric (MT) data. They also 3.1. Gravity survey design modelled the basement structure and inferred that the Waihi faults were pathways for magma intrusion into the TgVM. Johnson et al. We collected gravity data along radial traverses on foot, from the (2011) and Johnson and Savage (2012) used seismic anisotropy mea- summit of the volcano massif (~2200 m) to the tree line (~1100 m) surements to map spatial and temporal changes in anisotropy; in partic- where thick vegetation prevented further surveying. This results in a ular they found a strong change in anisotropy north of Mt Ngauruhoe 2–3 km wide region with no coverage from 1100 m to 700 m. Below which they associated with the TgVM hydrothermal system. Hill et al. 700 m, surveying resumed along the roads at the base of the volcano. (2015) undertook a detailed 3D MT survey and found evidence for The area with no coverage consists mostly of distal lava flows and pyro- both shallow and deep conductive zones, interpreted as magma ascent clastic deposits. Station spacing along the traverses is 500 m and tra- pathways and deeper storage zones. In particular, they found a narrow verses were located approximately 1 km apart; spacing reflects a (1 km) vertical conductive zone under Mt Ngauruhoe interpreted to trade-off between completing coverage of the entire volcano and reso- represent the ascent path of magmatic fluids from a source at 4-12 km lution of structures in the volcano. At 500 m station spacing, Nyquist depth. theorem indicates we will be able to resolve features with a wavelength of N1000 m which is considered adequate for the size of the volcano, but 2.2. Physical property measurements we would not be able to resolve small scale features such as individual feeder dykes as observed at Red Crater (Wadsworth et al., 2015). The TgVM consists of a variety of rock types including alternating However, we can resolve large scale fault offsets and bulk rock physical layers of highly vesicular scoria and dense lavas, underlain by dense property distributions related to different parts of the volcano and meta-sediments. To constrain the physical properties of different rock hydrothermal system. The station density across the survey area is units for modelling, we extracted a dataset of 176 samples from the 0.8 stations/km2 which increases to 1.4 stations/km2 on the upper GNS Science PetLab database (http://pet.gns.cri.nz)fromrockson flanks of the volcano. and around the TgVM. Physical properties include wet and dry density and magnetic susceptibility. We also incorporated physical property 3.2. Gravity datasets measurements from several studies of individual vents of the TgVM; including Tongariro (Hackett, 1985), North Crater (Griffin, 2007), Mt Our gravity dataset contains data from three sources. The oldest Ngauruhoe, (Krippner, 2009; Sanders, 2010), and Blue Lake (Simons, data, sourced from the GNS Science New Zealand gravity station 2014), for a total of 288 measurements. For analysis purposes, we database (http://gns.cri.nz/Home/Products/Databases/New-Zealand-

Table 1 Summary of physical properties for rock types within the TgVM. All volcanic includes all andesite lava and pyroclastic samples. Number of samples of each rock type is given by n.

Dry density Dry density Wet density Wet density Magnetic susceptibility Magnetic susceptibility

Mean (kg/m3) Stdev (kg/m3) Mean (kg/m3) Stdev (kg/m3) Mean (SI) Stdev (SI)

Andesite lava 2384 (n = 132) 302 2535 (n = 108) 205 0.022 (n = 94) 0.017 Pyroclastic 1591 (n = 143) 569 1931 (n = 143) 267 0.009 (n = 46) 0.010 All volcanic 1971 (n = 275) 607 2334 (n = 251) 364 0.017 (n = 140) 0.016 Greywacke 2706 (n = 31) 56 2727 (n = 30) 45 Sandstone 2407 (n = 3) 6 2517 (n = 3) 6 C.A. Miller, G. Williams-Jones / Journal of Volcanology and Geothermal Research 319 (2016) 12–28 15 16 C.A. Miller, G. Williams-Jones / Journal of Volcanology and Geothermal Research 319 (2016) 12–28

Gravity-Station-Network), provides absolute gravity values referenced ellipsoid. Our study area contains a wide variety of volcanic and base- to the IGSN71 gravity datum (Morelli et al., 1974). These data provide ment rock types above this datum, so finding a single density suitable far field, regional coverage to a distance of 40 km from the TgVM. for all rock types is not possible and may have resulted in parts of the Many of the 576 stations selected date from the 1960s, '70s and '80s dataset being over or under corrected. However we consider our chosen and were located with accuracies of 100 m horizontally and N5mverti- correction density to be in the middle of the range of all rock types, thus cally. Repeat occupation of a subset of these stations during the current any error caused by over or underestimation of the correction density survey reveals no systematic offset between current and old readings should be evenly distributed around the chosen value and not overly with most stations being repeatable to within 0.2 mGal. As a control bias the results. on data quality, we excluded GNS Science data with measured elevations that are grossly different (10s of m) from a 10 m Digital Elevation 3.4. Complete Bouguer anomaly Model (DEM). These mostly occur in areas of steep terrain to the east of the study area where the poor horizontal positioning results in a large We computed the complete Bouguer anomaly (CBA) on the dataset elevation difference. of 957 stations, covering an area of 70 km by 80 km in order to accurate- The second dataset comprises 66 stations on the massif from Cassidy ly determine the regional gravity in the area of interest around the and Locke (1995) and Cassidy et al. (2009). These data were located volcanoes (Fig. 2A). The regional CBA ranges from +41 mGal in the using a mixture of barometry, precise levelling and differential GPS, northwest to −54 mGal in the south and broadly consists of two gravity b hence height accuracies vary from ~5 m to 0.5 m. highs to the northwest and east. These highs correlate with mapped fi The third dataset is new data we collected at 315 stations over 2 eld areas of outcropping basement Torlesse Terrane in the east and the campaigns in 2014 and 2015. Vertical and horizontal positions were de- Waipapa Terrane in the west (Fig. 1). Between these highs is a broad termined using differential GNSS (using a Trimble Geoexplorer XH), op- gravity low defining the width of the Taupo Volcanic Zone. The TgVM erating in rapid static mode with 2 min occupations and post processed is situated at a local maximum in this gravity low which decreases fi fi with Trimble Path nder Of ce software using nearby GeoNet CGPS further to the north-east and also to the south, towards Mt Ruapehu. stations as reference stations. The short baselines between rover and The strike of the gravity signal changes from NE–SW to E–W just to b base stations ( 10 km) allows gravity stations to be located with verti- the south of Mt Ruapehu, representing the termination of the TVZ. cal accuracies of better than 0.2 m. In the modelling area (dashed box in Fig. 2A), the CBA shows a To combine the data from the 3 surveys, we reoccupied the primary broad, asymmetric ‘U-shaped’ trend, descending steeply from a gravity base station from the Cassidy survey and tied it to our newly established high (28 mGal) in the north-west to a broad low (−4 mGal) and then local base station. Repeat measurement of a selection of the Cassidy sta- increasing gradually to another high (12 mGal) in the east (Fig. 2B). tions showed values agreed within 0.12 mGal. Finally, we tied our local The TgVM is located on the west side of the gravity low, and is base station back to the National Park reference station (GNS station ID characterised by short wavelength anomalies with localised maxima 96), so that both our and the Cassidy stations were assigned an absolute and minima (±6 mGal). Local minima are associated with the gravity value, consistent with the GNS dataset. cones of Mt Ngauruhoe, Red Crater, Mt Tongariro summit and Te Maari as well as with the Upper Tama Lake crater. Local maxima are 3.3. Gravity data reduction and errors located in the Oturere Valley to the east of Red Crater and to the east of Blue Lake. A small gravity high is also located in the Mangatepopo We corrected the raw data from the 2014 and 2015 surveys for Earth Valley. fi tide, ocean loading and drift (e.g. Battaglia et al., 2012) to produce data Modelling requires removal of a regional eld that creates a long fl relative to our local base. Average base station loop closure errors after wavelength gradient across the anomaly map re ecting the broad crust- Earth tide and ocean loading are accounted for are 0.02 ± 0.03 mGal. al structure relating to the TVZ and the subduction zone to the east. Zeng fi fi We applied a correction scheme following that outlined in Hinze et al. and Ingham (1993) and Cassidy et al. (2009) ttheregional eld in the (2005), across all generations of data, to compute a consistent dataset. Tongariro area using a third order polynomial calculated from stations Details of the correction scheme are in supplementary materials. located on outcropping basement rocks. The use of a third order polyno- Estimating the overall error in gravity values from closure, height, mial is common in gravity studies throughout the TVZ (e.g., Stern, 1979; positioning and terrain errors gives a RMS value of 0.070 mGal for the Stagpoole and Bibby, 1999; Caratori Tontini et al., 2015). We remove the fi 2014 and 2015 surveys. The Cassidy survey data has errors from 0.1 to same regional eld from our data and the resulting residual anomaly 1 mGal, while errors from the GNS Science dataset are up to ~1 mGal, map is shown in Fig. 2B. mostly due to the poor accuracy of the height determination. One of the most important choices in gravity data reduction is the 3.5. Aeromagnetic data acquisition and processing selection of the reduction density applied to the Bouguer and terrain corrections. The shape and amplitude of the resulting complete Bouguer Approximately 510 line kilometres of aeromagnetic data were ac- anomaly can vary with choice of reduction density which directly influ- quired in February 1995, along 19 flight lines, at a nominal 500 m spac- ences the resulting models. Methods such as Nettleton (1939) and ing, of which only the southernmost line is published in Cassidy et al. Parasnis (1966) or their derivatives (Gottsmann et al., 2008), are often (2009). A proton precession magnetometer was towed 100 m behind not valid in heterogeneous volcanic rock environments so we use our a Cessna fixed-wing aircraft and data were acquired at 2 s intervals physical property dataset instead. See supplementary materials for fur- which for an average flight speed of 100 knots resulted in 1 sample ap- ther discussion on the calculation of correction density from gravity proximately every 100 m. The survey was flown at a constant altitude measurements. We chose the mean volcanic rock wet density value of configuration with a mean altitude of 2450 m, although turbulence 2334 kg/m3 (rounded down to 2300 kg/m3), to represent the bulk den- meant that flight altitude could vary by as much as 100 m above or sity of the volcano massif, for computing the complete Bouguer anoma- below the mean, along each line. Flight lines were oriented NW–SE ly. This density is valid for the mass above the reduction datum, i.e., the perpendicular to the main strike of the TVZ. A single tie line was flown

Fig. 2. Complete Bouguer anomaly data for A) regional area around Mt Tongariro, contour interval 5 mGal. The detailed 28 × 18 km model area is shown in the black dashed rectangle. Black dots are GNS Science stations, blue dots are Cassidy stations, red dots are stations collected in this study. B) The residual CBA in the model area after removal of a 3rd order polynomial, contour interval 2 mGal. Vent locations are shown in white triangles and stations as for part A. C) The residual CBA low pass filtered to 10,000 m wavelength, contour interval 2 mGal. Shown in all figures are the active faults (white lines). Coordinates are easting and northing in m using the NZTM projection. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.) C.A. Miller, G. Williams-Jones / Journal of Volcanology and Geothermal Research 319 (2016) 12–28 17

Fig. 3. A) Total magnetic intensity anomaly, contour interval 50 nT. Green dots show ﬂight lines. B) Reduced to pole (RTP) map of the TMI anomaly, contour interval 50 nT. Vent locations are overlain as white triangles, active faults overlain as white lines. Coordinates are easting and northing in m using the NZTM projection. (For interpretation of the references to color in this ﬁgure legend, the reader is referred to the web version of this article.)

NE–SE along the central axis of Mt Tongariro (Fig. 3A). We did not level 3.6. Residual Total Magnetic Intensity anomaly the survey using crossover points or the tie line due to difficulties in maintaining a constant altitude over the mountainous terrain. The residual Total Magnetic Intensity (TMI) anomaly (Fig. 3A) shows A base station installed 20 km from the centre of the survey area pro- an elongated ellipsoid of magnetic material, with the long axis oriented vided diurnal corrections. To produce a total magnetic intensity (TMI) parallel to the regional strike of the TVZ. The high flight altitude, at up to anomaly, we subtracted the International Geomagnetic Reference 1 km above the topography, results in a relatively low resolution image Field (IGRF) (Thébault et al., 2015) from each sample point. To remove but still offers enough detail to distinguish larger scale structures. The spikes in the data, we implemented a smoothing filter using a zero relatively young age of the TgVM means the volcanics were erupted phase, 11 sample Hanning window (Jones et al., 2001). We removed a within the current normally magnetised Brunhes epoch that began weak regional trend (~2 nT/km) from the smoothed TMI data by ~0.78 Ma and as such, no reversely magnetised material is expected subtracting a second order polynomial, creating a residual TMI dataset and magnetic lows imply loss of magnetisation. To aid qualitative inter- suitable for inversion. For all magnetic models we used an ambient pretation, we computed the reduced to pole (RTP) anomaly (Fig. 3B) to field with intensity 55,458 nT, inclination −64.58° and declination centre the anomalies over their causative body, but for inversion, the 20.67° as calculated from the IGRF model for February 1995. residual TMI data is used. 18 C.A. Miller, G. Williams-Jones / Journal of Volcanology and Geothermal Research 319 (2016) 12–28

The amplitude of the residual TMI reaches a maximum of ~1300 nT at Mt Ngauruhoe, a significant proportion of which is expected to be caused by the topographic effect of the high standing cone in relation to the flight altitude. In areas away from Mt Ngauruhoe, the field intensity averages around 250 nT and fades to 0 nT on the flanks of the volcano. The magnetisation reaches background levels while still over mapped lava exposures, however this may be a result of the high flight height reducing the sensitivity of the measurements on the lower flank. Within the high intensity zone in the centre of the volcanic massif is a 9km2 area of very low magnetisation. This area is centred between Red Crater and Blue Lake and extends south to Mt Ngauruhoe and north to Upper Te Maari Crater, coincident with the hydrothermal surface features. A ridge of magnetic high extends south of Mt Ngauruhoe over Upper and Lower Tama Lakes, while a small magnetic low is observed around the area of the Ketetahi hot springs on the northwest flank of the volcano. A weak positive anomaly is seen to the north of the Pukeonake cones, associated with the lava field from those cones (Fig. 3).

3.7. Spectral analysis

In order to investigate the internal structure of the TgVM, and its re- lationship to the basement, we apply a low pass filter to separate the potential field signals into long wavelengths representing the deeper basement and shorter wavelengths representing shallower volcanic material. To determine the optimal filter characteristics, we computed a radially averaged power spectrum from a 2D fourier transform of the residual CBA gravity data and TMI data using the GMT function, grdfft (Wessel and Smith, 2013). Depth to the top of the source can be calculated by decomposing the radially averaged power spectra into linear segments where the depth is proportional to the slope of line segments (Spector, 1970). For TMI data, the source depth is calculated from depth = −s/4π where s is the slope of the natural log spectral power (SP) vs wave number (k) graph. For gravity data, a correction term, 2 ∗ ln(k), is added to the ln(SP) to convert the gravity data to pseudomagnetic data (Hinze et al., 2005), before calculating the source depth using the same formula as for TMI data. We use an elevation of 1500 m for the average topographic height to convert gravity source depths to elevations and use 2450 m (flight height) as the reference Fig. 4. A) Radially averaged power spectrum of complete Bouguer anomaly data. Both for converting TMI source depths to elevations. Note, however, that corrected (dots) and non-corrected data (triangles) are shown. Top of source elevation the wavelength filtering is still based on the uncorrected power spec- estimates are based on the corrected data, while wavelength filter characteristics are based on the non-corrected data. Vertical lines highlight the wavelength segments trum, only the depth calculation for gravity data requires the correction. fi fi tested in ltering. B) Radially averaged power spectrum of TMI data with top of source The wavelength lter cut off was chosen at slope changes in the uncor- elevation estimates. Annotated line segments represent elevations of source layers. rected power spectrum. The corrected radially averaged gravity power spectra (Fig. 4A) is decomposed into 2 linear segments each representing a different source 4. Geological modelling and geophysical inversion depth within the data. The top of the shallowest source is equivalent to the topography of the volcanic material, while the top of a second We constructed a range of models, starting with simple uncon- source at around sea level likely corresponds to the top of the strained apparent property models, followed by unconstrained 3D greywacke basement. In the uncorrected data three wavelength seg- inversions, and finally geologically constrained 3D inversions. The ments are seen; longer than 10,000 m, 10,000 m to 3300 m and less workflow and subsequent interpretation based on building models of than 3300 m. We applied low pass filters of 10,000 m and 3300 m and increasing complexity allows the full dynamic range of the dataset to compared the resulting grids. The 3300 m filtered grid contained short be explored. wavelength anomalies that matched those visible in the unfiltered All modelling in this study uses GOCAD® Mining Suite (www. dataset and are spatially related to known vents, so we consider these mirageoscience.com) to construct the starting 3D geological model anomalies to be volcano related, rather than the basement. We there- which is directly coupled to the VPmg (Vertical Prism Magnetics fore applied a low pass Butterworth filter with a cut off of 10,000 m to Gravity) inversion routines (Fullagar and Pears, 2007; Fullagar et al., the gravity data in order to model the basement beneath the TgVM as 2008), providing two-way interaction between geology and geophysi- shown in Fig. 2C. cal data. VPmg allows for a variety of inversion types: Homogeneous The radially averaged power spectrum of the residual TMI data is property inversion, to determine the optimal physical property (density shown in Fig. 4B. The deepest layer is around 50 m below sea level or magnetic susceptibility) of a single geologic unit; Heterogeneous and likely represents the base of volcanic material as the basement is property inversion, to find the optimal physical property distribution non-magnetic. This interface corresponds well with a basement source within a unit. This includes an apparent property inversion, where the depth calculated from the gravity data. Shallower layers represent the voxet (a 3D regular grid-set of voxels, or volume-pixels) consists of a bulk of the volcanic material and a layer of surface lava probably associ- single vertical prism extending the full depth of the voxet. The misfit ated with high elevation material on Mt Ngauruhoe. of the apparent property inversion is useful for quickly assessing the C.A. Miller, G. Williams-Jones / Journal of Volcanology and Geothermal Research 319 (2016) 12–28 19 degree of three dimensionality in the data. Finally, geometry inversion faults based on the model of Cassidy et al. (2009) and consistent with of geological contacts optimise the shape of a unit while its physical the mapped continuation of the faults outside the study area by property remains constant. Each type of inversion can be applied se- Villamor and Berryman (2006b). We then warped the basement surface quentially and in combination. VPmg models the subsurface as a set of to fit the fault offsets. As a sensitivity test of our model to the starting vertical rectangular prisms whose top surface matches the topography. geometry, we also created a flat basement surface that only included Internally, the prisms are divided into cells with arbitrary vertical the outcropping basement, i.e., with no fault steps in it. dimension. Cell subdivisions can be based on geologic units and each unit can be assigned homogeneous or heterogeneous physical proper- 5. Results ties. Heterogeneous units can be inverted in a smooth sense via least squares or stochastically. In stochastic inversion, random perturbations We began our exploration of the data by first performing apparent are chosen for each cell of each geologic unit. The size of the random property inversions to determine the broad lateral distribution of phys- perturbations is governed by the a priori defined property distribution ical properties. We then performed an unconstrained 3D inversion to in- and limited to three standard deviations from the mean property vestigate the approximate vertical extent of anomalous features. These value of the unit being inverted. The perturbation is accepted if it re- results (see supplementary material) highlight the necessity to better duces the chi-squared misfit, and is rejected otherwise (Fullagar and constrain the depth to the basement interface which from our knowl- Pears, 2007). The RMS misfit (mGal) is computed and recorded, where edge of the geology and petrophysical contrasts suggests should be a vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi u sharp interface. u XN t 2 RMS ¼ 1=N ðÞOn−Cn ð1Þ n¼1 5.1. 3D geologically constrained density inversion

where N is the number of data, On is the measured data and Cn the To begin the geologically constrained model, we first used a geome- calculated model response. try inversion to adjust the shape of the starting basement surface to fit Mathematical details of the inversion method are provided in the low pass filtered gravity data. In the geometry inversion, the base- supplementary material. ment density contrast is fixed at 400 kg/m3 (to represent an absolute value of 2700 kg/m3 matching our petrophysical data) and the top of 4.1. Model initialisation the basement in each voxet cell is allowed to vary vertically. The resulting RMS misfit for this model is 1.7 mGal. The areas of worst misfit Voxet cell sizes are 250 m (half gravity station spacing) in the east are associated with older GNS Science stations that may have errors up and north axes and 100 m in the vertical (depth) axis. VPmg mathemat- to 1 mGal. To test for any density variations in the basement which may ically extends the model volume to 25 km depth to ensure complete improve the fit of the model, we performed a second inversion on the modelling of the data at all wavelengths. The voxet is then embedded starting basement surface, comprising a combined geometry and het- in a halfspace so that the model does not terminate abruptly, reducing erogeneous property inversion. In this approach, alternating steps of a edge effects. The physical property of the halfspace is optimised by the single heterogeneous density inversion iteration, followed by a single inversion routine. geometry inversion iteration are run to produce a model that accounts We gridded the observed gravity and magnetic data at a 250 m cell for both the geometry and density contrasts within the basement. This size to ensure that each vertical prism in the voxet is associated with 1 improved the misfit RMS to 0.75 mGal. data point located in the centre of the prism. VPmg requires the input With the shape of the basement now constrained, we model the of a topographic surface as the top surface of the model, so that the cover unit using a heterogeneous density inversion. In this model, the topography is modelled directly. This was constructed using a point best fit basement geometry and physical property distribution, as de- dataset from an 8 m DEM (down sampled to 24 m) as constraints for scribed above, is fixed and the initially homogeneous cover unit is con- fitting a smooth surface using the DSI interpolator in GOCAD. The verted to a heterogeneous unit. We then perform both a conventional resulting surface consists of a mesh of equilateral triangles with inversion and a stochastic inversion in the cover unit using the unfil- ~100 m sides. This topographic surface is then down sampled to the tered, full wavelength, gravity data. In this way we are fitting the 250 m voxet for modelling. remainder of the gravity signal, not accounted for by the basement We begin with a simple two layer case, consisting of a basement unit model, by density variations in the cover unit. The final conventional and a cover unit of volcanics; the complex geological history of the inversion model of the full dataset has an RMS of 0.9 mGal while the sto- TgVM makes constructing a more detailed model highly subjective. In- chastic inversion produces a model with RMS of 0.93 mGal, with the stead, we use the inversion process to discover detail within the subsur- highest misfits associated with poorer quality gravity stations. This mis- face and focus on building geological constraints into the model from fit is due to the combination of the fit of the basement surface plus the kilometre scale features. misfit due to the heterogeneous cover model. When the basement mis- To construct the basement surface, we imported into GOCAD shape fit is subtracted, the misfit due to the cover unit alone is 0.15 mGal. files of surface fault traces from the GNS Science New Zealand Active We extracted a series of elevation slices from the conventional and Fault Database and outlines of geological units from GNS Science stochastic inversions at 1350, 1150, 750 and 550 m a.s.l. (Fig. 5). The Hawke's Bay QMAP (Lee et al., 2011). We used the mapped contacts conventional and stochastic inversions show similar results, however, to accurately define the basement–volcanics boundary in our model. the stochastic model appears noisier as each cell is treated independent- We have not explicitly modelled the thin (~100 m) overlying layer of ly of its neighbour so no smoothing occurs. Tertiary sediments as they have a similar density to the average volcanic A general NE/SW trending fabric is evident, parallel to the main TVZ rock density and distinguishing the two without other constraints is dif- structural trend. A low density root occurs under Mt Ngauruhoe extend- ficult. Using the DEM and the basement geology contact curves, we ing from surface to the basement and a small low density root occurs be- warped a flat starting surface to the shape of the outcropping basement neath the recent Tama Lakes crater. A complex shaped low density body topography. We then overlaid the surface fault traces of the National occurs under Red Crater, Central Crater, North Crater and Blue Lake and Park, Waihi, Poutu and Rangipo fault zones. These fault zones are appears to terminate approximately at the western most Waihi fault. made from numerous sub-parallel strands and modelling each individ- This body, coincident with the known hydrothermal surface features, ual strand is outside the resolution of our gravity data. As such, only the is divided into two sub-areas by a narrow region of higher density major fault strands were included. We assigned initial offsets to these material. To the west of the massif are several surficial, low density 20 C.A. Miller, G. Williams-Jones / Journal of Volcanology and Geothermal Research 319 (2016) 12–28

Fig. 5. Depth slices from the conventional (left column B–E) and stochastic (right column G–J) geologically constrained gravity inversions at 1350, 1150, 750, 550 m a.s.l. Residual anomaly maps are shown in A and F. Overlain in black lines are active faults and vent locations as white triangles. Coordinates are easting and northing in m using the NZTM projection. anomalies which may be small pockets of pyroclastic material infilling is modelled at the head of the Oturere Valley, just to the east of Red Cra- previous topographic lows. ter. This is likely to represent a thick sequence of overlapping lava flows Small high density features exist from surface to a few hundred me- from Red Crater. Small pockets of dense rock occur north and south of tres depth. The largest of these is associated with the Te Tatu lavas from Mt Ngauruhoe and may represent accumulation of lavas from Mt Ngau- the recent NE Oturere vent (Fig. 5). A small area of high density material ruhoe. Dense rock to the south of Mt Ngauruhoe may also be related to C.A. Miller, G. Williams-Jones / Journal of Volcanology and Geothermal Research 319 (2016) 12–28 21 thick lavas from ancestral cones Tama 1 and Tama 2. An area of above 6. Discussion average density rock occurs south of Pukekaikiore around the Waihi fault, while the Poutu fault zone sits within lower than average density 6.1. Basement structure rocks. High density rocks in the far northwest and southeast correlate with outcropping and shallow basement. Our gravity model shows that the basement forms a continuous, but faulted, surface beneath the TgVM. The best fitting basement model includes a subtle E to W density gradient across the basement surface 5.2. 3D geologically constrained susceptibility inversion from 2670 kg/m3 to 2730 kg/m3 (Fig. 7). This is within the range of physical property measurements and likely reflects the change from To construct the magnetic model, we imported the basement unit Torlesse Terrane to Waipapa Terrane, respectively. McNamara et al. from the best fit gravity model and assigned it a zero susceptibility, (2014) noted a systematic difference in the mechanical behaviour of consistent with basement physical properties. We then performed con- the Torlesse and Waipapa Terranes, where the Waipapa Terrane ap- ventional and stochastic unconstrained inversions on the cover layer. pears to be mechanically stronger. They attributed this difference to The RMS of the conventional inversion is 7 nT and shows a normal dis- Waipapa greywacke having coarser grain size and a more maficcompo- tribution of residuals indicating no bias in the model. The stochastic in- sition compared to the finer grained, felsic composition of the Torlesse. version RMS is 49 nT and the inversion did not successfully converge. These attributes would also make the Waipapa Terrane rocks more We therefore do not use the stochastic inversion results further in our dense than the Torlesse, which corroborates our model and justifies study. the inclusion of a variable density basement. We are not able to distin- Fig. 6 shows depth slices from the conventional inversion. A guish any sharp boundary between the two terranes, under the TgVM. large very low susceptibility area (b0.02 SI) is imaged in the shallowest The basement is mostly flat-lying with a slight dip (b5°) to the north slices and is coincident with the surface manifestations of the and south away from the TgVM. Throws across the National Park, Waihi hydrothermal system and the low density body in the gravity model. 1 and Waihi 2, and Poutu faults are modelled as ~50 m, ~70 m, ~200– It is elongated in a NE–SW direction, sub-parallel to the main faults 300 m, and ~200 m, respectively. There is no discernible offset on the and is divided into two sub-areas by a narrow ridge of moderate Rangipo fault in the model area which has decreased from ~800 m in susceptibility rock which is not present in deeper slices. This area the south. The throws on these faults are similar to those obtained by extends to basement and is bound to the west by the Waihi fault. A Cassidy et al. (2009), however they did not model the Poutu fault. small shallow region of low susceptibility is associated with Ketetahi While we have modelled the main fault strands, the Waihi and Poutu hot springs. fault Zones are made up of numerous sub-parallel strands which may Areas of moderate magnetic susceptibility (0.05 SI) form the flanks accommodate the fault movement in a more complex way than can of the TgVM coincident with flank lava flows. Areas of high magnetic be resolved by our gravity model. susceptibility (N0.075 SI) occur at Mt Ngauruhoe and Tama Lakes. At The gravity model shows that surfaces between the fault zones are Mt Ngauruhoe a thick high magnetic susceptibility area extends to gently dipping (b5°) towards the centre of the TgVM, indicating that around sea-level. This area has two side lobes, a thick one to the each block of basement between the faults may be tilted as well as north-west, and one with less vertical extent to the south-east of the being faulted. The total basement subsidence beneath the TgVM deter- cone. These features extend from surface and likely represent a combi- mined from the basement gravity model, taking into account discrete nation of the central conduit and flank lava flowsfromMtNgauruhoe. faulting and surface tilting, is 500–700 m. Two more high magnetic susceptibility units exist around the Tama Villamor and Berryman (2006a) infer initiation of faulting 554 ± Lake vents. These are 300–500 m thick, extend from surface, and are as- 323 ka. If we assume that the onset of extension across the Ruapehu gra- sociated with both the young Tama vents, and with lavas from the older ben was synchronous with the onset of volcanism, then a long term sub- Tama 1 and 2 cones. sidence rate from the gravity models of 1.8–2.5 mm/year since 275 ka is calculated. This is similar to the 4 ± 1 mm/year of subsidence Villamor and Berryman (2006a) calculated for the Ruapehu graben to the south 5.3. Limitations and sensitivity of geophysical models of the TgVM. The general good agreement of geophysically derived subsidence Our ability to determine subsurface structure of the TgVM using rates with those measured from fault outcrops shows that most, if not gravity and magnetics fundamentally relies on there being sufficient all, subsidence can be accounted for by fault movement. There is no contrast in the physical properties of rock types that make up the massif. need to invoke additional mechanisms for subsidence such as crustal At the TgVM there is a wide range of physical property values and flexing under the volcanic load, as has been observed for large stratovol- the basement is non-magnetic providing a good contrast with the over- canoes erupted into weak sediments (e.g., Concepción and Maderas in lying volcanics. However, when we model the depth extent of the Nicaragua; van Wyk de Vries and Borgia, 1996). The mechanical stiff- demagnetised volcanic area, the non-magnetic basement may not pro- ness of greywacke is sufficient to support the weight of volcanic materi- vide sufficient contrast to determine if demagnetised volcanic rocks al and any unsupported load is taken up by fault movement. We are extend into the basement. unable to determine the effect adding volcanic load has had on the The basement is also generally denser than the overlying volcanics, rate of fault movement throughout the lifetime of the TgVM, however, however, some of the less vesicular lavas have similar densities to our geophysically -derived subsidence rates are slightly less than the greywacke, so if the basement is directly overlain by a thick pile of geologically observed rates so it is expected that loading has had little lava, accurately modelling the location of the basement contact will be impact on subsidence rate which is largely a result of tectonic extension. more difficult. Similarly, density contrasts between solidified intrusions 3D gravity and magnetic inversions did not resolve any discrete bod- and greywacke are likely to be low (b100 kg/m3) making them difficult ies within the basement, which is not surprising considering that there to detect within the basement using gravity. However, they should pro- is minimal density contrast between solidified andesite and greywacke vide good magnetic targets if they are of suitable size and have not been and the size and depth of solidified magnetic intrusions may not be re- hydrothermally altered. Low density roots of magma conduits or low solvable by the aeromagnetic survey. Conversely, it suggests that any density hydrothermally altered rocks should have good contrast within partially molten, low density magma bodies are too small or too deep the basement. to be detected by our measurements. This is in agreement with the pet- End members of the range of constrained and unconstrained models rologic model of Hobden et al. (1999) that suggests small discrete are shown in supplementary material. magma batches are responsible for recent magmatic activity. 22 C.A. Miller, G. Williams-Jones / Journal of Volcanology and Geothermal Research 319 (2016) 12–28

6.2. Volcanic ediﬁce structure

We calculate a volume of ~350 km3 of volcanic material between the topographic surface and the basement. Most of this material is sourced from the TgVM, however, a small portion (estimated at b50 km3)willbe sourced from outside the TgVM, from the Taupo Caldera and from nearby Mt Ruapehu. We calculate a long-term volumetric eruption rate of 1.3 × 10−3 km3/year since initiation of volcanism at 275 ka. Hobden (1997) estimated a total volume of around 60 km3 for eruptive products since the initial Tama 1 eruptions, erupting at an average rate of 0.17 × 10−3 km3/year, while eruption rates for the Holocene formation of Mt Ngauruhoe are around 0.3 × 10−3 km3/year. However, Nairn (2000) and Pardo et al. (2012) provide evidence for much larger eruption rates of the TgVM (Pahoka–Mangamate sequence of 6 km3 in 200–400 years) and Mt Ruapehu (0.1 to 10 km3 from individual Plinian eruptions since 27 ka), suggesting that the rate of magma supply has been variable over time. The total eruptive volume and rate derived from our geophysical model is five to six times larger than those estimated from field observations and suggests a significant amount of material has been removed by erosion and glaciation, or that the volumes of older cones have been significantly underestimated because of poor surface exposure and a lack of knowledge about the true basement depth beneath the TgVM. These revised volumetric output rates compare well to the global average rate for andesite volcanoes of 2.3 ± 0.8 × 10−3 km3/year (White et al., 2006), and imply a greater rate of magma production, albeit with a large amount of temporal variability. Our model resolves some structures relating to the long term individual cone building episodes outlined by Hobden (1997),particularly the thick lavas associated with the Tama 1 and 2 centres, south of Mt Ngauruhoe. Features associated with younger volcanic vents (Fig. 8A) include a small low density volume coincident with the young Pukekaikiore cone and similar volumes mapped at Tama Lakes, Half Cone and the young NE Oturere vent. These low density features likely represent the tops of vesicular magma feeder systems. The lack of high density anomalies coincident with these vents suggests that magma drainage from the feeder post-eruption, as seen at Red Crater (Wadsworth et al., 2015), may be common at other vents. This in turn suggests these vents were erupted at a low magma supply rate, likely originating from a small discrete magma batch. The only high density anomalies are associated with mapped thick lava flows around old Pukekaikiore, at the head of the Oturere Valley and in the head of the Mangahouhounui Valley. A large low density and magnetic root (Fig. 8B) extends to below basement depth under Mt Ngauruhoe which we attribute to a substantial magma feeder system. This cone shaped feature is geographically and geophysically consistent with the low velocity zone imaged by Rowlands et al. (2005) and the low resistivity zone imaged by Hill et al. (2015). An extensive irregularly shaped low density area is mapped under the Red Crater, Central Crater and North Crater area, extending to Ketetahi hot springs to the north. This feature is broadly coincident with a low magnetic susceptibility volume and is interpreted as altered rock from the hydrothermal system. Hydrothermal systems can produce either high or low density alteration depending on whether void space is mineralised by circulating fluids and thereby increasing density, or if the host rock is altered to clay minerals (Allis, 1990). As the hydrothermal system at the TgVM is vapour dominated (Hochstein, 1985), there is not likely to be large fluid circulation precipitating minerals into void space. Alteration of rocks by hot acid gases and condensate, into lower density clay minerals is therefore the most likely mechanism

Fig. 6. Depth slices from the conventional geologically constrained TMI inversion at 1350, 1150, 750, 550 m a.s.l. (B–E). Residual anomaly map shown in A. Overlain in black lines are active faults and vent locations as white triangles. Coordinates are easting and northing in m using the NZTM projection. C.A. Miller, G. Williams-Jones / Journal of Volcanology and Geothermal Research 319 (2016) 12–28 23

Fig. 7. Parallel view from south showing the top of greywacke basement surface colour coded by density. The topography has been raised above the basement surface for clarity. No vertical exaggeration. for lowering density. Other low density regions on the flanks of the observed by Johnson and Savage (2012), which they attributed to in- massif are shallow, have small vertical extent and are interpreted as var- creased fracturing associated with the hydrothermal reservoir. Such iations in accumulations of tephra. fracturing would help account for the observed gravity low in addition Our model provides limited support for the conclusion of Cassidy to the formation of low density clay minerals as a by-product of hydro- et al. (2009) that the basement faults acted as pathways for magma as- thermal alteration (Allis, 1990). Within this area are zones of more in- cent. While some high density and magnetic material exists in conjunc- tensely demagnetised rock (0.001 SI). These areas are on the south tion with the southern end of the Waihi fault, this is also coincident with flank of the Tongariro summit ridge, around Blue Lake and the slopes thick lavas from Pukekaikiore to the north of their profile. Our models above Upper Te Maari crater and coincide with areas of considerable do not image any thin, dense vertical structures extending into the base- surface alteration. A smaller area of demagnetised rock is associated ment. To be resolved by our surveys, feeder dykes would need to be a with Ketetahi hot springs where there is an extensive area of surface al- minimum of several hundred metres wide which is geologically implau- teration. Ballistic blocks from the 2012 Upper Te Maari eruption show sible based on comparison with outcropping dykes and the amount varying degrees of alteration, from fresh to extensively altered (Breard of extension across the faults required to be accommodated by dyke et al., 2014) while mineral component analysis by Pardo et al. (2014) infilling during a single eruptive episode. The other fault systems found unaltered magnetite phenocrysts within the ejecta. We interpret around TgVM show no evidence of dyke intrusion so it appears that these observations as being consistent with Te Maari's location on the while these faults are important tectonic features, they are not critical edge of the hydrothermal system where a mixture of fresh and altered in determining the location of eruption sites. It may be that the faults rock occurs. are not zones of weakness but rather, are strongly coupled and resistant As the basement is non-magnetic, it is difficult to determine how far to dyke intrusion. into the basement the hydrothermal system extends solely on the basis Neither the gravity nor the magnetic models imaged any large intru- of the magnetic anomaly. The geologically constrained model suggests sive bodies within the edifice. This is in contrast to the dense dyke net- that the low magnetisation zone ends approximately 200 m above the work imaged at the Pouakai and Kaitake volcanoes in Taranaki, to the basement surface while the unconstrained model suggests it may con- west of the TVZ (Locke et al., 1993). The extensive Taranaki volcano tinue to 2500 m below sea level (see supplementary materials). The feeder systems suggest a fundamentally different magma supply regime low density (2250 kg/m3) region coincident with the low magnetisation compared to the TgVM and again corroborates petrophysical evidence area, extends to 500 m below the basement in the unconstrained model, that the TgVM vents are feed from small discrete magma batches that suggesting that alteration might occur in the basement rock to a depth have risen from depth with little intermediate storage. of a few hundred metres. This is not surprising given that the basement is likely to be highly fractured from repeat eruptions through it, 6.3. Hydrothermal system allowing fluids to more easily circulate in otherwise impermeable rocks (cf. the basement-hosted Kawerau geothermal field, Milicich We interpret that the large magnetic low and the complex region of et al., 2013). All models show that the hydrothermal system appears low density north of Mt Ngauruhoe extending to Upper Te Maari repre- to truncate against the Waihi faults in the west. Faulted low permeabil- sents the extent of TgVM hydrothermal system (Fig. 9). The total vol- ity basement rocks may form a seal for the hydrothermal system to the ume of hydrothermally altered rock is around 20 km3. This volume is west which agrees with the earlier observation that faults are tightly smaller than the approximately 100 km3 estimated by Caratori Tontini coupled and impermeable to fluid or magma injection. This may also ex- et al. (2010) for the hydrothermal system at Marsili volcano in the plain why there are no outflow hot springs or other thermal features Tyrrhenian Sea, but is significantly larger than the 1.5–3km3 estimated outside these faults. by Finn et al. (2001, 2007), for the volumes of altered rock at Mt Rainier The southern boundary of the hydrothermal system ends on the and Mt Adams, Washington. The low density and demagnetised area is north side of Mt Ngauruhoe. This appears unusual given that Mt Ngau- broadly consistent with a region of strong seismic anisotropy change ruhoe has been the dominant cone building centre during the last 7 ka 24 C.A. Miller, G. Williams-Jones / Journal of Volcanology and Geothermal Research 319 (2016) 12–28

Fig. 8. A) Iso-surfaces of low density (2250 kg/m3) in cyan and high density (2400 kg/m3) in red. B) Iso-surface of low magnetic susceptibility (0.025 SI) in yellow and high magnetic susceptibility (0.09 SI) in purple. Perspective view looking from the south-east. No vertical exaggeration. (For interpretation of the references to color in this ﬁgure legend, the reader is referred to the web version of this article.) and would be expected to have a well developed hydrothermal system. masking of altered rocks seems unlikely. Secondly, it is possible that Several explanations are possible. Firstly, the hydrothermal system does either the present Mt Ngauruhoe cone, or remnants of older cones extend under Mt Ngauruhoe, but demagnetised areas are masked by the (Tama 1 and 2) act as a physical barrier to ﬂuid movement. However, strongly magnetised historic lavas. Our modelling shows that strongly the current Mt Ngauruhoe cone is low density implying high porosity magnetised rocks exist to basement depth under Mt Ngauruhoe, so rocks that should be capable of supporting a hydrothermal system and

Fig. 9. The TgVM hydrothermal system as shown by iso-surfaces of low magnetic susceptibility (0.025 SI) in yellow, and low density (2250 kg/m3) in cyan. The area outlined in red represents the hydrothermal system as shown in Fig. 10. Perspective view looking from the east south-east. No vertical exaggeration. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.) C.A. Miller, G. Williams-Jones / Journal of Volcanology and Geothermal Research 319 (2016) 12–28 25 seismic evidence suggests hydrothermal fluid movement beneath the small landslide of ~0.0007 km3 (Procter et al., 2014). Both landslide de- cone (Jolly et al., 2012). Alternatively, it may be that Mt Ngauruhoe is posits show extensive evidence of weak hydrothermally altered materi- simply too young to have been sufficiently hydrothermally altered to al being a contributing factor in the failure. The volume of alteration at produce a measurable demagnetisation. Estimating the rate of rock the TgVM is considerably larger than that found at other studied andes- dissolutioninhydrothermalsystemsisdifficult as rates of chemical re- ite volcanoes and constitutes a considerable potential hazard. Our actions that drive dissolution are highly dependent on temperature, sur- gravity and magnetic models suggest that highly altered surface zones face area of exposed rock and fluid flux rates through the rock. Caratori surround a core of more moderately altered rock that extends to base- Tontini et al. (2015) estimated a dissolution rate of 50,000 m3/year for ment depths. Similarly, variably altered cores have been mapped at the Rotomahana geothermal field based on comparison of rates for other andesite volcanoes (Finn et al., 2001; Mayer et al., 2015) and are other hydrothermal systems. For instance at Poás volcano, Rowe et al. recognised as potential sources of future landslides. As alteration ex- (1992) calculated a rate of ~1650 m3/year, and at White Island, tends to basement depths at the TgVM, the risk of large scale flank col- Giggenbach (1987) estimated 22,000 m3/year of rock dissolution. The lapse is increased compared to volcanoes with only shallow alteration. topographic volume of the Mt Ngauruhoe cone is approximately Alteration at depth is susceptible to failure by mechanisms generated 2.2 km3 which at a dissolution rate of 50,000 m3/year would require within the volcanic edifice, such as dyke intrusion or changes in 44,000 years to demagnetise. Dissolution rates would need to be an pressurisation of the hydrothermal system, as well as surface based pro- order of magnitude higher, for sufficient alteration to have developed cesses. In addition, once surface initiated flank collapse is under way, within the ~7 ka life span of the cone. Hence, while it is likely that the there is no unaltered core of competent rock to impede collapse retro- hydrothermal system does extend under Mt Ngauruhoe, the rocks gression, and limit the amount of material available to form debris there are simply too young to have been sufficiently demagnetised to flows. be imaged by aeromagnetic surveys. We can apply the same argument To determine to a first order, landslide risk areas on the TgVM, we in reverse to conclude that the hydrothermal system producing the undertook a slope angle analysis of the DEM to identify steep slopes co- large demagnetisated area to the north of Mt Ngauruhoe must be incident with altered rock, which are likely to be susceptible to failure long-lived, on the order of 104 to 105 years. (Fig. 10). Steep slopes (N30°), coincident with regions of hydrothermal alteration, are at greater risk of failure (Moon et al., 2005) due to the 6.4. Implications for volcanic hazards inferred from geophysical models lower friction properties of altered rocks. Slope failure often depends on the level of the groundwater system, or on the internal fluid pore Our geophysical models provide first order constraints on volcanic pressure, both of which can be raised through injection of fluids as a re- hazard potential at TgVM and offer some possibility for improvements sult of dyke intrusion, such as occurred in 2012. A suitable trigger mech- in hazard monitoring. Use of simple 1D seismic velocity models in anism may then be small local earthquakes or shaking from a larger volcanic earthquake location algorithms can impact real-time regional event or increased rainfall adding extra gravitational load hazard assessment in times of volcanic unrest if those models are (Voight and Elsworth, 1997). oversimplified, resulting in incorrect hypocenter locations. We compare The areas of coincident steep slope and hydrothermal alteration our 3D density model to the 1D velocity model calculated by Jolly et al. occur on slopes of all aspects and size. The most at risk slopes, and (2014) for an area on the north flanks of Te Maari. Converting P-wave those representing the highest hazard, are high on the massif where velocity (Vp) to density (Brocher, 2005)wefind excellent agreement they have large potential energy and the longest runouts. Such slopes below 1 km depth between densities in the 3D gravity model and the occur on the north flanks above Upper Te Maari and on the west flanks Vp converted densities (Table 2). Our gravity model resolution is less of Tongariro summit, above the Mangatepopo Valley and around the sensitive to changes in the top 1 km which may account for the discrep- head of the Mangahouhounui and Oturere Valleys. A more detailed ancy in the upper layer. We could therefore reasonably convert our 3D ground study of alteration and slope stability in these areas would pro- density model to a volcano wide high resolution 3D velocity model for vide a better assessment of the hazard these slopes present. improved earthquake locations. This would be especially useful for high frequency earthquakes within the volcanic edifice (Hurst et al., 7. Conclusions 2014), with seismic wavelengths short enough to be influenced by a heterogeneous seismic velocity distribution. Geologically constrained geophysical inversions of an extensive po- Our basement faulting model shows little evidence for the Waihi and tential field data set at the TgVM have successfully mapped the base- Poutu faults acting as magmatic pathways and suggests that magma in- ment structure beneath the volcano, identified magmatic plumbing trudes the basement between the faults, rather than along them. These system roots, delineated the extent of a large hydrothermal system, fault structures are therefore considered low probability areas for future highlighted areas at risk from various volcanic hazards and offer im- eruption locations. provements to volcanic unrest monitoring capability. The delineation of the extensive hydrothermal system identifies Our model shows a continuous dense, non-magnetic basement areas most at risk of phreatic eruption, either as a result of natural beneath the volcano, and suggests places where it is pierced by the mag- fluctuations in hydrothermal activity or those perturbed by magmatic matic plumbing system. The basement is extensively down faulted to a intrusion. While phreatic eruptions are possible wherever there is depth of around 100 m below sea level under the TgVM, a total to 500– interaction between magma and water, intrusions into the base of the 700 m displacement across the graben. We calculate that the volume of hydrothermal system have greater ability to provide early warning of volcanic material above the basement is five to six times larger than pre- future eruption than intrusions into shallow groundwater aquifers vious geologically based estimates, requiring a higher rate of magma (e.g., Hurst et al., 2014). The presence of large volumes of hydrothermally weakened rock further promotes the chances of phreatic activity due to lowered confining strengths of these rocks (Heap et al., 2015). The Table 2 large number of tourists that cross the TgVM on hiking trails each P wave velocities from Jolly et al. (2014) converted to density using the relationships in Brocher (2005). year, means that even small phreatic eruptions can present a high risk to those in close proximity. Depth 1D Vp Converted densities Gravity model densities 3 3 The TgVM has a history of landslides and flank collapse. The largest (km) (km/s) (kg/m ) (kg/m ) example is the 0.5 km3 Te Whaiau formation, formed by collapse of 0 1.8 1810 2200 the northwestern flank around 55–60 ka (Lecointre et al., 2002) while 1 3.6 2330 2334 2 5.9 2690 2700 the most recent example is the 2012 Te Maari eruption, initiated by a 26 C.A. Miller, G. Williams-Jones / Journal of Volcanology and Geothermal Research 319 (2016) 12–28

Fig. 10. Slope angle (calculated on 15 m DEM) and extent of demagnetised hydrothermal system (outlined in red). Areas of greatest risk of collapse are steep slopes within the hydrothermal system outline. Coordinates are easting and northing in m using the NZTM projection. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.) supply than previously thought. However, the lack of discrete magma Natalia Deligne, Sophie Pearson, Alex Kmoch, Tom Ayling, Janvion bodies within the model indicates that when magma is supplied to Cevuard, Matt Stott and Nick MacDonald. Thanks to Vaughan Stagpoole the surface it is only via relatively small batches. We find only minor ev- for running the terrain corrections on the GNS system. Thanks to Harry idence for the Waihi faults acting as preferential magma pathways for Keys at the Department of Conservation for permitting assistance and to the Pukekaikiore vents and conclude that if feeder dykes have intruded helicopter pilots Keith McKenzie and Andrew MacIntosh for logistical along them, they are likely too small to be resolved by our surveys. Thus, support. Thomas Campagne, Peter Fullagar, Stanislawa Hickey and the main bounding faults are considered low probability areas for future Shannon Frey at Mira Geoscience provided many hours of VPmg and eruptions. We image a low density and highly magnetic root beneath GOCAD assistance. Thanks to Jeff Zurek and Jeff Witter for discussions Mt Ngauruhoe which we interpret to represent the main system of on the models and to Bruce Christenson, Dougal Townsend and Graham magmatic feeder conduits beneath the volcano. Leonard for discussions on the hydrothermal system and geology of the The zone of hydrothermally demagnetised rocks extends to around TgVM. We thank Fabio Caratori-Tontini and an anonymous reviewer for basement depths and is bound to the west by the Waihi fault system, comments that improved the manuscript. Figures and analysis were which acts as an impermeable barrier to fluid movement. The hydro- made using open source software, Python, Matplotlib (Hunter, 2007), thermal system is not imaged beneath Mt Ngauruhoe, so while there Inkscape and QGIS. is other evidence for hydrothermal activity there, fluid circulation with- fi in the cone is likely not developed enough to cause suf cient alteration Appendix A. Supplementary data to be detected by our survey. Where the hydrothermal system inter- fl sects steep topographic slopes, we map areas most at risk from ank Supplementary data to this article can be found online at http://dx. collapse. Flank collapse potential may be elevated by pressurisation doi.org/10.1016/j.jvolgeores.2016.03.012. of the hydrothermal system following dyke intrusion, or by over- saturation of pore space during heavy rainfall or snow melt. Finally, References we propose that our high resolution density model could act as a proxy for a new 3D velocity model to improve earthquake locations Acocella, V., Spinks, K., Cole, J., Nicol, A., 2003. Oblique back arc rifting of Taupo Volcanic and enhance volcanic unrest monitoring capability. Zone, New Zealand. Tectonics 22. http://dx.doi.org/10.1029/2002TC001447. Allis, R., 1990. Geophysical anomalies over epithermal systems. J. Geochem. Explor. 36, 339–374. http://dx.doi.org/10.1016/0375-6742(90)90060-N (URL: http://www. Acknowledgements sciencedirect.com/science/article/pii/037567429090060N). Battaglia, M., Poland, M.P., Kauahikaua, J., 2012. GTools: an interactive computer program Corinne Locke and John Cassidy are thanked for their generous ac- to process gravity data for high resolution applications. AGU Fall Meeting. Beetham, R.D., Watters, W.A., 1985. Geology of Torlesse and Waipapa terrane basement cess to unpublished gravity and aeromagnetic data. C.M. is supported rocks encountered during the Tongariro Power Development project, North Island, by the Earthquake Commission (EQC), the New Zealand Ministry of New Zealand. N. Z. J. Geol. Geophys. 28, 575–594. http://dx.doi.org/10.1080/ Business, Innovation and Employment (MBIE) Core funding to GNS 00288306.1985.10422534. fi Blaikie, T., Ailleres, L., Betts, P., Cas, R., 2014. A geophysical comparison of the Science, Mira Geoscience and Mitacs Accelerate Canada. Many eld as- diatremes of simple and complex maar volcanoes, Newer Volcanics Province, sistants helped with gravity data collection including Nellie Olsen, south-eastern Australia. J. Volcanol. Geotherm. Res 276, 64–81. http://dx.doi. C.A. Miller, G. Williams-Jones / Journal of Volcanology and Geothermal Research 319 (2016) 12–28 27

org/10.1016/j.jvolgeores.2014.03.001 (URL: http://linkinghub.elsevier.com/ Hobden, B.J., Houghton, B.F., Davidson, J.P., Weaver, S.D., 1999. Small and short-lived retrieve/pii/S0377027314000717). magma batches at composite volcanoes: time windows at Tongariro volcano, New Breard, E., Lube, G., Cronin, S., Fitzgerald, R., Kennedy, B., Scheu, B., Montanaro, C., White, Zealand. J. Geol. Soc. 156, 865–868. http://dx.doi.org/10.1144/gsjgs.156.5.0865. J., Tost, M., Procter, J., Moebis, A., 2014. Using the spatial distribution and lithology of Hochstein, M., 1985. Steaming Ground at Red Crater and in the Te Maari Craters. ballistic blocks to interpret eruption sequence and dynamics: August 6, 2012 Upper Hunt, T., Mumme, T., 1986. Some magnetic properties of two recent lava flows from Te Maari eruption, New Zealand. J. Volcanol. Geotherm. Res. 286, 373–386. http:// Mount Ngauruhoe, Tongariro Volcanic Centre, New Zealand. J. R. Soc. N. Z. 16, dx.doi.org/10.1016/j.jvolgeores.2014.03.006 (URL: http://linkinghub.elsevier.com/ 71–83. http://dx.doi.org/10.1080/03036758.1986.10426957. retrieve/pii/S0377027314000936). Hunter, J., 2007. Matplotlib: a 2D graphics environment. Comput. Sci. Eng. 9, 90–95. Brocher, T.M., 2005. Empirical relations between elastic wavespeeds and density in the Hurst, T., Jolly, A.D., Sherburn, S., 2014. Precursory characteristics of the seismicity before the Earth's crust. Bull. Seismol. Soc. Am. 95, 2081–2092. http://dx.doi.org/10.1785/ 6 August 2012 eruption of Tongariro volcano, North Island, New Zealand. J. Volcanol. 0120050077. Geotherm. Res. 286, 294–302. http://dx.doi.org/10.1016/j.jvolgeores.2014.03.004 Cannavò, F., Camacho, A.G., González, P.J., Mattia, M., Puglisi, G., Fernández, J., 2015. Real (URL: http://linkinghub.elsevier.com/retrieve/pii/S0377027314000821). time tracking of magmatic intrusions by means of ground deformation modeling Johnson, J.H., Savage, M.K., 2012. Tracking volcanic and geothermal activity in the Tongariro during volcanic crises. Sci. Rep. 5, 10970. http://dx.doi.org/10.1038/srep10970. Volcanic Centre, New Zealand, with shear wave splitting tomography. J. Volcanol. Caratori Tontini, F., Cocchi, L., Muccini, F., Carmisciano, C., Marani, M., Bonatti, E., Ligi, M., Geotherm. Res. 223-224, 1–10. http://dx.doi.org/10.1016/j.jvolgeores.2012.01.017 Boschi, E., 2010. Potential-field modeling of collapse-prone submarine volcanoes in (URL: http://linkinghub.elsevier.com/retrieve/pii/S0377027312000303). the southern Tyrrhenian Sea (Italy). Geophys. Res. Lett. 37. http://dx.doi.org/10. Johnson, J.H., Savage, M.K., Townend, J., 2011. Distinguishing between stress-induced and 1029/2009GL041757 (n/a–n/a). structural anisotropy at Mount Ruapehu volcano, New Zealand. J. Geophys. Res. 116, Caratori Tontini, F., de Ronde, C., Scott, B., Soengkono, S., Stagpoole, V., Timm, C., Tivey, M., B12303. http://dx.doi.org/10.1029/2011JB008308. 2015. Interpretation of gravity and magnetic anomalies at Lake Rotomahana: geolog- Jolly, A.D., Chardot, L., Neuberg, J., Fournier, N., Scott, B.J., Sherburn, S., 2012. High impact ical and hydrothermal implications. J. Volcanol. Geotherm. Res. http://dx.doi.org/10. mass drops from helicopter: a new active seismic source method applied in an active 1016/j.jvolgeores.2015.07.002 (URL: http://linkinghub.elsevier.com/retrieve/pii/ volcanic setting. Geophys. Res. Lett. 39. http://dx.doi.org/10.1029/2012GL051880 S0377027315002103). (n/a–n/a). Cassidy, J., Locke, C.a., 1995. Geophysical signatures of andesite volcanoes in New Zealand — Jolly, A., Lokmer, I., Kennedy, B., Keys, H., Proctor, J., Lyons, J., Jolly, G., 2014. Active seismic contrasts and structural implications. sources as a proxy for seismic surface processes: an example from the 2012 Tongariro Cassidy, J., Ingham, M., Locke, C.A., Bibby, H., 2009. Subsurface structure across the axis of the volcanic eruptions, New Zealand. J. Volcanol. Geotherm. Res. 286, 317–330. http://dx. Tongariro Volcanic Centre, New Zealand. J. Volcanol. Geotherm. Res. 179, 233–240. doi.org/10.1016/j.jvolgeores.2014.04.008 (URL: http://linkinghub.elsevier.com/ http://dx.doi.org/10.1016/j.jvolgeores.2008.11.017 (URL: http://linkinghub.elsevier. retrieve/pii/S0377027314001188). com/retrieve/pii/S0377027308006100). Jones, E., Oliphant, T.E., Peterson, P., 2001. Scipy: Open Source Scientific Tools for Python. Christenson, B.W., Werner, C.A., Reyes, A.G., Sherburn, S., Scott, B.J., Miller, C., Rosenburg, M.J., Krippner, J.B., 2009. Ngauruhoe Inner Crater Volcanic Processes of the 1954–55 and Hurst, A.W., Britten, K.A., 2007. Hazards from hydrothermally sealed volcanic conduits. 1974–75 Eruptions (Ph.D. thesis) University of Waikato. Eos, Trans. Am. Geophys. Union 88, 53. http://dx.doi.org/10.1029/2007EO050002. Lecointre, J.A., Neall, V.E., Wallace, C.R., Prebble, W.M., 2002. The 55- to 60-ka Te Whaiau Day, S.J., 1996. Hydrothermal pore fluid pressure and the stability of porous, permeable Formation: a catastrophic, avalanche-induced, cohesive debris-flow deposit from volcanoes. Geol. Soc. Lond. Spec. Publ. 110, 77–93. http://dx.doi.org/10.1144/GSL.SP. Proto-Tongariro Volcano, New Zealand. Bull. Volcanol. 63, 509–525. http://dx.doi. 1996.110.01.06 (URL: http://sp.lyellcollection.org/content/110/1/77). org/10.1007/s004450100167. Finn, C., Sisson, T., Deszcz-Pan, M., 2001. Aerogeophysical measurements of collapse-prone Lee, J., Townsend, D., Bland, K., Kamp, P., 2011. Geology of the Hawkes Bay Area. Technical hydrothermally altered zones at Mount Rainier volcano. Nature 409, 600–603. http:// Report. GNS Science. dx.doi.org/10.1038/35054533 (URL: http://www.ncbi.nlm.nih.gov/pubmed/11214315). Locke, C.A., Cassidy, J., MacDonald, A., 1993. Three-dimensional structure of relict strato- Finn, C.A., Deszcz-Pan, M., Anderson, E.D., John, D.A., 2007. Three-dimensional geophysi- volcanoes in Taranaki, New Zealand: evidence from gravity data. J. Volcanol. cal mapping of rock alteration and water content at Mount Adams, Washington: Geotherm. Res. 59, 121–130. http://dx.doi.org/10.1016/0377-0273(93)90081-2 implications for lahar hazards. J. Geophys. Res. 112, B10204. http://dx.doi.org/10. (URL: http://linkinghub.elsevier.com/retrieve/pii/0377027393900812). 1029/2006JB004783. López, D.L., Williams, S.N., 1993. Catastrophic volcanic collapse: relation to hydrothermal Fullagar, P., Pears, G., 2007. Towards geologically realistic inversion. Proceedings of Explo- processes. Science (New York, N.Y.) 260, 1794–1796. http://dx.doi.org/10.1126/ ration 07: Fifth Decennial International Conference on Mineral Exploration, science.260.5115.1794. pp. 444–460 (URL: http://www.dmec.ca/ex07-dvd/E07/pdfs/28.pdf). Mayer, K., Scheu, B., Gilg, H.A., Heap, M.J., Kennedy, B.M., Lavallée, Y., Letham-Brake, M., Fullagar, P.K., Pears, G.A., McMonnies, B., 2008. Constrained inversion of geologic Dingwell, D.B., 2015. Experimental constraints on phreatic eruption processes at surfaces pushing the boundaries. Lead. Edge 27, 98–105. http://dx.doi.org/10.1190/ Whakaari (White Island volcano). J. Volcanol. Geotherm. Res. 302, 150–162. http:// 1.2831686. dx.doi.org/10.1016/j.jvolgeores.2015.06.014 (URL: http://linkinghub.elsevier.com/ Giggenbach, W.F., 1987. Redox processes governing the chemistry of fumarolic gas dis- retrieve/pii/S0377027315001900). charges from White Island, New Zealand. Appl. Geochem. 2, 143–161. http://dx.doi. McNamara, D., Faulkner, D., McCarney, E., 2014. Rock properties of greywacke basement org/10.1016/0883-2927(87)90030-8. hosting geothermal reservoirs, New Zealand: preliminary results. Thirty-Ninth Work- Gottsmann, J., Camacho, A.G., Mart, J., Wooller, L., Fernández, J., Garca, A., Rymer, H., 2008. shop on Geothermal Reservoir Engineering, pp. 1–10. Shallow structure beneath the Central Volcanic Complex of Tenerife from new gravity Milicich, S., Wilson, C., Bignall, G., Pezaro, B., Bardsley, C., 2013. Reconstructing the geological data: implications for its evolution and recent reactivation. Phys. Earth Planet. Inter. and structural history of an active geothermal field: a case study from New Zealand. 168, 212–230. http://dx.doi.org/10.1016/j.pepi.2008.06.020 (URL: http://linkinghub. J. Volcanol. Geotherm. Res. 262, 7–24. http://dx.doi.org/10.1016/j.jvolgeores.2013.06. elsevier.com/retrieve/pii/S003192010800160X). 004 (URL: http://linkinghub.elsevier.com/retrieve/pii/S0377027313001789). Griffin, A.M., 2007. Products and Processes of Cone-building Eruptions from North Crater, Moebis, A., Cronin, S.J., Neall, V.E., Smith, I.E., 2011. Unravelling a complex volcanic history Tongariro (Ph.D. thesis) University of Waikato. from fine-grained, intricate Holocene ash sequences at the Tongariro Volcanic Centre, Hackett, W.R., 1985. Geology and Petrology of Ruapehu and Related Vents (Ph.D. thesis) New Zealand. Quat. Int. 246, 352–363. http://dx.doi.org/10.1016/j.quaint.2011.05.035 Victoria University of Wellington. (URL: http://linkinghub.elsevier.com/retrieve/pii/S1040618211003089). Hagerty, M., Benites, R., 2003. Tornillos beneath Tongariro Volcano, New Zealand. J. Volcanol. Moon, V., Bradshaw, J., Smith, R., de Lange, W., 2005. Geotechnical characterisation of Geotherm. Res. 125, 151–169. http://dx.doi.org/10.1016/S0377-0273(03)00094-5 stratocone crater wall sequences, White Island Volcano, New Zealand. Eng. Geol. (URL: http://linkinghub.elsevier.com/retrieve/pii/S0377027303000945). 81, 146–178. http://dx.doi.org/10.1016/j.enggeo.2005.07.014. Heap, M.J., Kennedy, B.M., Pernin, N., Jacquemard, L., Baud, P., Farquharson, J.I., Scheu, B., Morelli,C.,Gantar,C.,Honkasalo,T.,McConnel,R.,Tanner,J.,Szabo,B.,Uotila,U.,Whalen,C., Lavallée, Y., Gilg, H.A., Letham-Brake, M., Mayer, K., Jolly, A.D., Reuschlé, T., Dingwell, 1974. The International Gravity Standardisation Net 1971 (IGSN71). Special Publication D.B., 2015. Mechanical behaviour and failure modes in the Whakaari (White Island No. 4. volcano) hydrothermal system, New Zealand. J. Volcanol. Geotherm. Res. 295, Nairn, I.A., 2000. The; 10 ka multiple vent pyroclastic eruption sequence at Tongariro 26–42. http://dx.doi.org/10.1016/j.jvolgeores.2015.02.012 (URL: http://linkinghub. Volcanic Centre, Taupo Volcanic Zone, New Zealand: part 1. Eruptive processes elsevier.com/retrieve/pii/S037702731500058X). during regional extension. J. Volcanol. Geotherm. Res. 19–44. Hill,G.J.,Bibby,H.M.,Ogawa,Y.,Wallin,E.L.,Bennie,S.L.,Caldwell,T.G.,Keys,H., Nettleton, L.L., 1939. Determination of density for reduction of gravimeter observations. Bertrand, E.A., Heise, W., 2015. Structure of the Tongariro Volcanic system: Geophysics 4. http://dx.doi.org/10.1190/1.1437088. insights from magnetotelluric imaging. Earth Planet. Sci. Lett. 432, 115–125. Parasnis, D., 1966. Mining Geophysics. Elsevier, Amsterdam–London–New York. http://dx.doi.org/10.1016/j.epsl.2015.10.003 (URL: http://linkinghub.elsevier. Pardo, N., Cronin, S., Palmer, A., Procter, J., Smith, I., 2012. Andesitic Plinian eruptions at com/retrieve/pii/S0012821X15006305). Mt. Ruapehu: quantifying the uppermost limits of eruptive parameters. Bull. Hinze, W.J., Aiken, C., Brozena, J., Coakley, B., Dater, D., Flanagan, G., Forsberg, R., Volcanol. 74, 1161–1185. http://dx.doi.org/10.1007/s00445-012-0588-y. Hildenbrand, T., Keller, G.R., Kellogg, J., Kucks, R., Li, X., Mainville, A., Morin, R., Pardo, N., Cronin, S.J., Németh, K., Brenna, M., Schipper, C.I., Breard, E., White, J.D., Procter, Pilkington, M., Plouff, D., Ravat, D., Roman, D., Urrutia-Fucugauchi, J., Veronneau, M., J., Stewart, B., Agustn-Flores, J., Moebis, A., Zernack, A., Kereszturi, G., Lube, G., Auer, Webring, M., Winester, D., 2005. New standards for reducing gravity data: the A., Neall, V., Wallace, C., 2014. Perils in distinguishing phreatic from phreatomagmatic North American gravity database. Geophysics 70, J25–J32. http://dx.doi.org/10. ash; insights into the eruption mechanisms of the 6 August 2012 Mt. Tongariro erup- 1190/1.1988183 (URL: http://scholarcommons.sc.edu/geol_facpub/2). tion, New Zealand. J. Volcanol. Geotherm. Res. 286, 397–414. http://dx.doi.org/10. Hobden, B.J., 1997. Modelling Magmatic Trends in Time and Space: Eruptive and Magmatic 1016/j.jvolgeores.2014.05.001 (URL: http://linkinghub.elsevier.com/retrieve/pii/ History of the Tongariro Volcanic Complex, New Zealand (Ph.D. thesis) University of S037702731400136X). Waikato. Potter, S.H., Jolly, G.E., Neall, V.E., Johnston, D.M., Scott, B.J., 2014. Communicating the sta- Hobden, B.J., Houghton, B.F., Lanphere, M.A., Nairn, I.A., 1996. Growth of the Tongariro tus of volcanic activity: revising New Zealands volcanic alert level system. J. Appl. volcanic complex: new evidence from KAr age determinations. N. Z. J. Geol. Geophys. Volcanol. 3, 13. http://dx.doi.org/10.1186/s13617-014-0013-7 (URL: http://www. 39, 151–154. http://dx.doi.org/10.1080/00288306.1996.9514701. appliedvolc.com/content/3/1/13). 28 C.A. Miller, G. Williams-Jones / Journal of Volcanology and Geothermal Research 319 (2016) 12–28

Procter, J., Cronin, S., Zernack, A.V., Lube, G., Stewart, R., Nemeth, K., Keys, H., 2014. Debris Thébault, E., Finlay, C.C., Beggan, C.D., Alken, P., Aubert, J., Barrois, O., Bertrand, F., Bondar, flow evolution and the activation of an explosive hydrothermal system; Te Maari, T., Boness, A., Brocco, L., Canet, E., Chambodut, A., Chulliat, A., Cosson, P., Civet, F., Du, Tongariro, New Zealand. J. Volcanol. Geotherm. Res. 286, 303–316. http://dx.doi. A., Fournier, A., Fratter, I., Gillet, N., Hamilton, B., Hamoudi, M., Hulot, G., Jager, T., org/10.1016/j.jvolgeores.2014.07.006 (URL: http://linkinghub.elsevier.com/retrieve/ Korte, M., Kuang, W., Lalanne, X., Langlais, B., Léger, J.M., Lesur, V., Lowes, F.J., pii/S0377027314002170). Macmillan, S., Mandea, M., Manoj, C., Maus, S., Olsen, N., Petrov, V., Ridley, V., Reid, M.E., Sisson, T.W., Brien, D.L., 2002. Volcano collapse promoted by hydrothermal Rother, M., Sabaka, T.J., Saturnino, D., Schachtschneider, R., Sirol, O., Tangborn, A., alteration and edifice shape, Mount Rainier, Washington. Geology 29, 779–782. Thomson, A., Tøffner-Clausen, L., Vigneron, P., Wardinski, I., Zvereva, T., 2015. Interna- http://dx.doi.org/10.1130/0091-7613(2001)029b0779:VCPBHAN2.0.CO;2. tional geomagnetic reference field: the 12th generation. Earth, Planets Space 67, 79. Rowe, G.L., Brantley, S.L., Fernandez, M., Fernandez, J.F., Borgia, A., Barquero, J., 1992. http://dx.doi.org/10.1186/s40623-015-0228-9 (URL: http://www.earth-planets- Fluid–volcano interaction in an active stratovolcano: the crater lake system of Poás space.com/content/67/1/79). volcano, Costa Rica. J. Volcanol. Geotherm. Res. 49, 23–51. http://dx.doi.org/10. Tontini, F.C., de Ronde, C.E.J., Kinsey, J.C., Soule, A., Yoerger, D., Cocchi, L., 2013. Geophys- 1016/0377-0273(92)90003-V. ical modeling of collapse-prone zones at Rumble III seamount, southern Pacific Rowland, J.V., Sibson, R.H., 2001. Extensional fault kinematics within the Taupo Volcanic Ocean, New Zealand. Geochem. Geophys. Geosyst. 14, 4667–4680. http://dx.doi.org/ Zone, New Zealand: softlinked segmentation of a continental rift system. N. Z. 10.1002/ggge.20278. J. Geol. Geophys. 44, 271–283. http://dx.doi.org/10.1080/00288306.2001.9514938. van Wyk de Vries, B., Borgia, A., 1996. The role of basement in volcano deformation. Geol. Rowlands, D.P., White, R.S., Haines, A.J., 2005. Seismic tomography of the Tongariro Volca- Soc. Lond. Spec. Publ. 110, 95–110. http://dx.doi.org/10.1144/GSL.SP.1996.110.01.07. nic Centre, New Zealand. Geophys. J. Int. 163, 1180–1194. http://dx.doi.org/10.1111/j. Villamor, P., Berryman, K.R., 2006a. Evolution of the southern termination of the Taupo 1365-246X.2005.02716.x. Rift, New Zealand. N. Z. J. Geol. Geophys. 49, 23–37. http://dx.doi.org/10.1080/ Sanders, F., 2010. Rheology and Flow Emplacement Processes of the 1954 Lavas, Mount 00288306.2006.9515145. Ngauruhoe Ph.D. thesis University of Waikato. Villamor, P., Berryman, K.R., 2006b. Late Quaternary geometry and kinematics of faults at Sano, Y., Kagoshima, T., Takahata, N., Nishio, Y., Roulleau, E., Pinti, D.L., Fischer, T.P., 2015. the southern termination of the Taupo Volcanic Zone, New Zealand. N. Z. J. Geol. Ten-year helium anomaly prior to the 2014 Mt. Ontake eruption. Sci. Rep. 5, 13069. Geophys. 49, 1–21. http://dx.doi.org/10.1080/00288306.2006.9515144. http://dx.doi.org/10.1038/srep13069. Voight, B., Elsworth, D., 1997. Failure of volcano slopes. http://dx.doi.org/10.1680/geot. Scott, B.J., Potter, S.H., 2014. Aspects of historical eruptive activity and volcanic unrest at 1997.47.1.1. Mt. Tongariro, New Zealand: 18462013. J. Volcanol. Geotherm. Res. 286, 263–276. Wadsworth, F.B., Kennedy, B.M., Branney, M.J., von Aulock, F.W., Lavallée, Y., Menendez, http://dx.doi.org/10.1016/j.jvolgeores.2014.04.003 (URL: http://linkinghub.elsevier. A., 2015. Exhumed conduit records magma ascent and drain-back during a com/retrieve/pii/S0377027314001127). Strombolian eruption at Tongariro volcano, New Zealand. Bull. Volcanol. 77, 71. Seebeck, H., Nicol, A., Villamor, P., Ristau, J., Pettinga, J., 2014. Structure and kinematics of http://dx.doi.org/10.1007/s00445-015-0962-7. the Taupo Rift, New Zealand. Tectonics 33, 1178–1199. http://dx.doi.org/10.1002/ Walsh, F., Hochstein, M., Bromley, C., 1998. The Tongariro Geothermal System. Review of 2014TC003569. Geophysical Data. Simons, B.C., 2014. Volcanic History and Eruption Processes of Blue Lake Crater, Tongariro Wessel, P., Smith, W., 2013. Generic mapping tools: inproved version released. EOS Trans. Ph.D. thesis University of Waikato. AGU 94, 409–410. Spector, A., 1970. Statistical Models for Interpreting Aeromagnetic Data. http://dx.doi. White, S., Crisp, J.A., Spera, F., 2006. Longterm volumetric eruption rates and magma org/10.1190/1.1440092. budgets. Geochem. Geophys. 7, Q03010. http://dx.doi.org/10.1029/2005GC001002. Stagpoole, V.M., Bibby, H.M., 1999. Residual gravity anomaly of the Taupo Volcanic Zone, Zeng, Y., Ingham, M., 1993. Modelling of gravity data from Tongariro National Park. 15th New Zealand, 1:250,000. Technical Report. Institute of Geological and Nuclear NZ Geothermal Workshop, pp. 219–226. Sciences. Stern, T.A., 1979. Regional and residual gravity fields, central North Island, New Zealand. N. Z. J. Geol. Geophys. 22, 479–485. http://dx.doi.org/10.1080/00288306.1979. 10424156. Supplementary material for “Internal structure and volcanic hazard potential of Mt Tongariro, New Zealand, from 3D geologically constrained gravity and magnetic models”

C. A. Millera,b,∗, G. Williams-Jonesa

aDepartment of Earth Sciences, Simon Fraser University , Burnaby, BC V5A 1S6, Canada bGNS Science, Wairakei Research Centre, Private Bag 2000, Taupo, 3352, New Zealand

1. Geophysical model

The geophysical model is supplied as a Geoscience Analyst project ﬁle for visualisation. The model ﬁle can be downloaded from http://tinyurl.com/ ntw5tel. Geoscience Analyst may be obtained free of charge from Mira Geoscience, http://www.mirageoscience.com/our-products/software-product/geoscience-analyst.

2. Gravity data correction scheme.

We corrected data to the ellipsoid height datum as our heights are largely derived from GNSS (WGS84 datum) measurements. Orthometric heights from the older surveys were converted to ellipsoid heights using a geoid model (NZVGD09) from Land Information New Zealand, so that they could be processed in a uniform manner with the new stations. We corrected all data for latitude variations using the theoretical gravity value from 1980 GRS Geodetic Reference System (GRS80) formula. As our survey covers a large range of elevations (600 to 2300 m.a.s.l.) we applied an atmospheric correction to account for the changing density of atmosphere. We applied the free air correction using the full equation

∗Corresponding author Email address: [email protected] (C. A. Miller)

Preprint submitted to Journal of Volcanology and Geothermal Research November 14, 2015 with second order terms and applied a Bouguer correction, with the curvature correction included. Finally we applied terrain corrections in a 3 step process. Firstly, inner terrain corrections out to Hammer Zone D were estimated in the field and then calculated using Hammer’s formula (Hammer, 1939). We then applied outer corrections using an 8 m DEM for Zones E and F and used a 15 m DEM for corrections in zones G to M. The terrain corrections represent one of the larger sources of error. Comparison of the manual inner (B-D) corrections with those from the 8 m DEM show a mean variation of 0.04 mGal. To estimate errors in the outer zones (E-M) we computed terrain corrections with an 8 m DEM and a 15 m DEM and calculate an RMS error in the differences between the two of 0.045 mGal. The DEMs used for terrain corrections do not take into account the bathymetry of lakes and as such the terrain corrections may be over-estimated due to low density water being corrected as denser rock (Hasegawa et al., 2009). Lake Taupo is New Zealand’s largest lake (616 km2) and is located 13 km north of the model area, but is within the broader area used to determine the regional field. To determine the gravity effect of the lake, we constructed a forward model using a slab with the approximate outline of the lake and a thickness of 100 m (the average lake depth). The maximum gravity effect at a station directly on the lake shore is 6 mGal, reducing to 0.02 mGal for the nearest station within our modelling area. We subtracted the gravity signal of Lake Taupo from our terrain correction value to ensure our stations used for regional field determination are properly corrected. The smaller Lake Rotoaira (13 km2), directly on the north edge of the model area, is on average 10 m deep and has a maximum gravity effect at the closest station of 0.023 mGal. This is within our noise envelope so we did not apply a further correction. No corrections were applied for the many small lakes on Mt Tongariro as they are too small to produce a measurable gravity effect.

2 3. Bulk Density from Gravity Measurements

The Nettleton (Nettleton, 1939) and Parasnis (Parasnis, 1966) methods provide means of estimating bulk density from minimisation of the correlation of Bouguer and terrain corrected gravity with topography and variants of these methods have been applied in several volcano studies (Gottsmann et al., 2008; Hautmann et al., 2013). We compare calculated bulk density estimations to rock physical property measurements to determine the optimal gravity correction density. An important and often overlooked caveat for both methods is that there is no direct causation of the topography by the geology, i.e., that dense rocks do not form high peaks or vice versa. This assumption can be difficult to verify in volcanic terranes where high standing ridges are often composed of denser, more erosion resistant material such as lava flows, while lower density pyroclastics fill valley floors. Both methods also rely on there being no regional affect in the gravity data which could bias the analysis. To minimise these re- strictions, we selected data from 156 stations located between the Waihi and Poutu fault zones where Cassidy et al. (2009) showed that the regional field is relatively flat. These data should be mostly free of long wavelength variations caused by large scale faulting and regional gravity changes, and best represent the bulk density of the volcano superstructure from 1300 to 2300 m elevation. For Nettleton’s method we calculated the Pearson correlation coefficient between the Bouguer gravity and elevation for a range of Bouguer and terrain correction densities. The correlation coefficient closest to zero, corresponds to the bulk terrain density of 2025 kg/m3 (Supplementary figure 1). We used the same data set for Paranis method and ran an ordinary least squares regression on the free air anomaly against the elevation data. The density is retrieved by the relation: density = regression slope / 2πG, where G is the universal gravity constant (Supplementary figure 2). All except two points plot within the 95% confidence interval of the best fit regression line. These points are located on the summit of Ngauruhoe and hand specimen densities from the 1954 summit cone are the lowest in the physical property dataset, between 498 - 731 kg/m3.

3 As these points are outside the 95% conﬁdence bounds, they are excluded from the regression analysis. The retrieved density is 2029 ± 36 kg/m3and agrees very well with the density retrieved from Nettleton’s method. The bulk density values retrieved by Nettleton’s and Parasnis’s methods are close to the mean wet density of pyroclastic materials sampled (1931 ± 267 kg/m3). However, this density appears to be too low as andesite lava (wet density 2535 ± 205kg/m3), makes up a sizeable proportion of the massif (Supplementary Figure 3. We interpret this discrepancy as due to a violation of the assumptions of Nettleton’s and Paranis’s method, implying that even with careful data selection there is still an implicit correlation between the topography and density. As such we use our physical property measurements for the correction density (Supplementary ﬁgure 3).

4. Inversion method

VPmg solves the inversion using the steepest descent method, where no ma- trix inversion is required (Fullagar and Pears, 2007). For the conventional least squares inversion, the objective at each iteration is the smallest parameter perturbation required to halve the L2 -norm data misfit. Stochastic inversion is an option for heterogeneous property inversion of either basement or other units. In this case, individual model cells are subjected to random property perturbations; the perturbation is accepted if it produces a reduction in misfit and if it is compatible with the expected property distribution within the geological unit. Maximum perturbation size is defined in terms of absolute property change, for property inversion, or in terms of fractional change in depth, for geometry inversion. Degree of fit is determined by the magnitude of the chi-squared data norm, L2, and the L1- data norm, defined by

N 2 X on − cn L2 = 1/N (1) ε n=1 n

N p X on − cn L1 = 1/N π/2 (2) ε n=1 n

4 where N is the number of data, On is the measured data, Cn the calculated model response and εn the uncertainty (standard deviation) applied to the nth data point. If the data uncertainties are controlled by normal random variables with zero mean, the both L2 and L1 have expected values of unity. Therefore, the model is deemed acceptable if L2 <1 and/or if L1 <1. For a starting εn, we used a value of 10% of the standard deviation of the range of observed data. If a model converges with this εn we successively lower the εn until the inversion stalls, at which point we have achieved the best ﬁt possible. The L1 and L2 misﬁts are dimensionless, as they are normalised by the uncertainties, εn. A depth weighting is applied to lower the sensitivity of the inversion to deeper cells, with the aim of creating ”smaller” bodies and limiting the smearing of bodies with depth.

5. Gravity results

5.1. Apparent Density Inversion

We begin our exploration of the gravity data by performing an ‘apparent density’ (AD) inversion. An AD model is a voxet with only 1 cell per vertical prism (i.e. the cell vertical extent is the same as the whole voxet) and the inversion adjusts the density in each prism until the calculated response of the model matches the input data. Apparent density models are useful for looking at lateral density variations in the dataset. The result of the apparent density inversion is seen in Supplementary Figure 4A. We use a diverging colour scheme to highlight areas of higher and lower density, relative to the Bouguer correction density of 2300 kg/m3. The model has an RMS misfit of 1.0 mGal which is relatively high and indicates that a more detailed model which incorporates vertical density variations is required . The residual gravity (observed - calculated) map (Supplementary Figure 4B) shows the highest areas of misfit are under the main part of the volcanic edifice, where the geology is likely to be highly three dimensional. Areas of high positive

5 misfit are also seen in this area, and also in some areas on basement rock. The basement areas contain data from stations which may have errors up to 1 mGal associated with them, and this could be reflected in the misfit in these areas. The AD model consists of a central low density zone flanked by high densities either side. The low density area is mostly confined between the Waihi and Poutu faults and the high density region correlates with outcropping or shallow basement. Within the central low density zone, several areas of higher or lower density exist. Two areas of high density occur, one to the south of Pukekaikiore and the other to the east of Blue Lake. An area of low density is associated with Mt Ngauruhoe and another low density area occurs around Red Crater.

5.2. 3D Unconstrained Density Inversion

To explore the depth extent of lateral density variations identified in the apparent density inversion we construct a model using a 3D unconstrained inversion. In this model we use the same voxet as the apparent density inversion, however we sub divide the vertical dimension into smaller cells. The top 2000 m of cells are constant thickness of 100 m and those deeper increase in thickness via an expansion factor of 1.1 times the previous cell thickness. The expanding cell size results in cells at -5000 m.b.s.l. that are ∼400 m thick reflecting the lower resolution of the data at depth. The voxet is filled initially with cells of 0 kg/m3 density contrast, equivalent to the bulk density of 2300 kg/m3 . We set the target RMS misfit to be 0.1 mGal (similar to our observation errors), to ensure we extract maximum information from our high quality data. The final model successfully converged with an RMS of 0.15 mGal. A representative depth slice of the resulting model voxet at 350 m.a.s.l. is shown in Supplementary Figure 5A, along with the residual misfit anomaly map and histogram in Supplementary Figure 5B. In this depth slice we see significantly more detail than from the AD inversion. High densities in the northwest and east correspond to outcropping basement. The broad low density zone between the Waihi and Poutu faults persists, as do the smaller high density zones to the east of Blue Lake and along the Waihi fault south of Pukekaikiore. Out-

6 side the central fault zone a more complex pattern of high and low densities occur. High density west of the Pukeonake vents correspond to the lava ﬁeld from these cones and high density along the north part of the Waihi fault corresponds to lava ﬂows from North Crater. High density east of the Poutu fault may relate to shallower basement east of this fault, however some small high and low density anomalies east of the Poutu fault are also associated with high residuals, so should be treated with caution. These areas are also associated with gravity stations that have higher errors (up to 1 mGal) than the rest of the dataset. The 3D unconstrained inversion model may have overestimated the thickness of many of the features in the model. For instance it is unlikely that the lavas from Pukeonake cones extend from ∼1100 m.a.s.l. at surface to 350 m.a.s.l.

6. Magnetic Results

6.1. Apparent Susceptibility Inversion

We initially invert the magnetic data for an apparent susceptibility model to show the broad lateral distribution of magnetic rocks. The result of this inversion is shown in Supplementary Figure 6A, along with a map and histogram showing the distribution of residuals from the inversion, Supplementary Fig- ure 6B). The RMS of the apparent susceptibility inversion is 8 nT, <1 % of the maximum anomaly amplitude. The main features are a pronounced area of very low to no susceptibility to the north of Mt Ngauruhoe, extending to Upper Te Maari Crater. This low is ringed by a series of magnetic highs with a maximum susceptibility on the north ﬂank of Mt Ngauruhoe. A small area of low susceptibility is coincident with the Ketetahi hotsprings and a broad area of moderate magnetic susceptiblity is located around the Tama Lakes.

6.2. 3D Unconstrained Susceptibility Inversion

We next performed a 3D unconstrained susceptibility inversion to deﬁne the lateral and vertical extent of both magnetic and non magnetic bodies within

7 the volcano. From this model we extract a representative of depth slice at 750 m.a.s.l. to show the main magnetic and non-magnetic features (Supplementary Figure 7A). The RMS of this inversion is 4 nT and shows a normal distribution of residuals indicating no bias in the model (Supplementary Figure 7B). To constrain the model to geologically realistic values the range of susceptibilities was allowed to vary between 0 and 0.2 SI. This is greater than our limited range of physical property measurements however magnetic susceptibility measurements on hand specimens are commonly under estimated compared to bulk rock composition, and are within ranges reported for similar sized andesite stratovolcanoes e.g. Finn et al. (2007). The dominant feature of this model is a very low to no magnetisation zone under the central part of the massif, extending to the north side of Ngauruhoe and to Upper Te Maari Crater. This zone is broadly coincident with the low density anomaly resolved from the gravity data. Surrounding this zone is a narrow irregular ring of higher magnetised rock, coincident with flank lava flows, extending out to the Waihi and Poutu faults. Within this ring an area of low magnetisation is coincident with the Ketetahi hotsprings. The area of highest magnetisation occurs at Mt Ngauruhoe and around the Upper Tama Lakes. An area of high magnetisation on the extreme south of the model relates to lava flows from Mt Ruapehu. West of the Waihi fault an area of low magnetisation occurs to the north of the Pukeonake cones, coincident with a gravity high, and may be related to lava flows from theses vents.

7. Limitations and Sensitivity of Geophysical Models

A way of testing the sensitivity of our geologically constrained models is to compare them to the unconstrained 3D models. If features introduced in our geologically constrained models are discernible in the unconstrained models then we can have a higher degree of conﬁdence in their validity. Likewise within the geologically constrained models we can test a range of starting models and see how they closely they converge on a single model through the inversion

8 process. In each case the RMS misfit of the model indicates all models are mathematically acceptable representations of the observed data. The sensitivity of features to a constrained basement depth is judged by if they extend below basement depths in the unconstrained model, or terminate at or above the basement in the geologically constrained model. As an example we compare the iso-surfaces of the low density root imaged beneath Ngauruhoe from the 3D unconstrained inversion and the geologically constrained model (Supple- mentary Figure 8A). In the unconstrained inversion the low density root from Mt Ngauruhoe appears to extend into the basement to a depth of -300 m. In the geologically constrained model this body is terminated at the basement, however the low pass filtered data used to construct the basement surface may have removed the signal of a small low density body close under the basement interface so it is not unreasonable to image the extension of the low density body into the basement in the unconstrained 3D model using the full wavelength dataset. This is also feasible when we consider that repeat magmatic intrusion is likely to leave the basement highly fractured, effectively lowering its density. Similarly we compare the depth extent of the low magnetic susceptibility body, associated with the hydrothermal system from unconstrained and geologically constrained models in Supplementary Figure 8B. In the geologically constrained model the low susceptibility zone terminates a few hundred metres above the basement, while in the unconstrained model a larger volume of very low susceptibility material extends into the basement. This example also illustrates the geophysical equivalence of a smaller higher susceptibility body and a larger, lower susceptibility body. The two examples may be taken as end members of the set of likely models. Finally we test the sensitivity of our geologically constrained inversion to the initial starting model used to seed the inversion. In the first model we constrained the basement surface to follow the outcrop topography in the Kaimanawa Ranges and under the TgVM made a flat lying surface at 300 m elevation. In the second model we used the mapped faults and prior information about their offsets to construct a geologically feasible starting surface. Geometry inversions

9 on both surfaces using a homogeneous basement density converged on similar results with the same RMS (1.7 mGal) misfit. While the inversion using the flat basement surface starting model did not explicitly recreate faulted offsets, the inverted surface does intersect the locations of the fault traces seen in the inversion of the faulted starting model. The final basement depths between the Poutu and Waihi fault zones agree within 100 m (∼10% of the total depth) in both models, showing that the retrieved shape of the basement surface is mostly insensitive to the initial starting model.

References

Cassidy, J., Ingham, M., Locke, C.A., Bibby, H., 2009. Subsurface structure across the axis of the Tongariro Volcanic Centre, New Zealand. Jour- nal of Volcanology and Geothermal Research 179, 233–240. URL: http: //linkinghub.elsevier.com/retrieve/pii/S0377027308006100, doi:10. 1016/j.jvolgeores.2008.11.017.

Finn, C.A., Deszcz-Pan, M., Anderson, E.D., John, D.A., 2007. Three- dimensional geophysical mapping of rock alteration and water content at Mount Adams, Washington: Implications for lahar hazards. Journal of Geophysical Research 112, B10204. URL: http://doi.wiley.com/10.1029/ 2006JB004783, doi:10.1029/2006JB004783.

Fullagar, P., Pears, G., 2007. Towards geologically realistic inversion. Proceed- ings of Exploration 07: Fifth Decennial International Conference on Mineral Exploration , 444–460URL: http://www.dmec.ca/ex07-dvd/E07/pdfs/28. pdf.

Gottsmann, J., A.G. Camacho, Mart´ı,J., Wooller, L., Fern´andez, J., Garc´ıa, A., Rymer, H., 2008. Shallow structure beneath the Central Volcanic Complex of Tenerife from new gravity data: Implications for its evolution and recent reactivation. Physics of the Earth and Planetary Interi-

10 ors 168, 212–230. URL: http://linkinghub.elsevier.com/retrieve/pii/ S003192010800160X, doi:10.1016/j.pepi.2008.06.020.

Hammer, S., 1939. Terrain Corrections for Gravimeter Stations. Geophysics 4, 184–194. URL: http://dx.doi.org/10.1190/1.1440495, doi:10.1190/ 1.1440495.

Hasegawa, T., Yamamoto, A., Kamiyama, H., Nakagawa, M., 2009. Gravity structure of Akan composite caldera, eastern Hokkaido, Japan: Application of lake water corrections. Earth, Planets and Space 61, 933–938. URL: http: //link.springer.com/10.1186/BF03353205, doi:10.1186/BF03353205.

Hautmann, S., Camacho, A.G., Gottsmann, J., Odbert, H.M., Syers, R.T., 2013. The shallow structure beneath Montserrat (West Indies) from new Bouguer gravity data. Geophysical Research Letters 40, 5113–5118. URL: http://doi.wiley.com/10.1002/grl.51003, doi:10.1002/grl.51003.

Nettleton, L.L., 1939. Determination of density for reduction of gravimeter observations. Geophysics 4. doi:10.1190/1.1437088.

Parasnis, D., 1966. Mining Geophysics. Elsevier, Amsterdam-London-New York.

11 Supplementary Figure 1: Plot of result of Nettleton’s method, showing density at the minimum Pearson correlation coeﬃcient between elevation and Bouguer corrected gravity for each density.

12 Supplementary Figure 2: Plot of result of Parasis method, showing density derived from the slope of the free air anomaly vs elevation.

13 Supplementary Figure 3: Histograms of physical property measurements for a range of rock types.

14 Supplementary Figure 4: A) Apparent density model. Location of vents in white triangles and active faults in black lines. B) Residual gravity distribution and histogram.

15 Supplementary Figure 5: A) Slice of unconstrained density inversion model at a 350 m.a.s.l. Location of vents in white triangles and active faults in black lines. B) Residual anomaly map and histogram of residuals, RMS = 0.15 mGal, from the unconstrained density inversion.

16 Supplementary Figure 6: A) Result of apparent susceptibility inversion. Location of vents in white triangles and active faults in black lines. B) Residual anomaly map and histogram of residuals, RMS = 8 nT, from the apparent susceptibility inversion.

17 Supplementary Figure 7: A) Result of unconstrained susceptibility inversion at a 750 m.a.s.l. Location of vents in white triangles and active faults in black lines. B) Residual anomaly map and histogram of residuals, RMS = 4 nT, from the unconstrained susceptibility inversion.

18 A Mt Ngauruhoe 2000 m Red Crateri S Maari Te N

1000 m

0 m

B 2000 m

0 m

-2500 m

0 10

Horizontal Distance (km)

Supplementary Figure 8: Perspective views from south-east showing: A) magnetic susceptibility iso-surfaces of the hydrothermal system extent from unconstrained inversion (yellow - 0.001 SI) and geologically constrained inversion (red - 0.025 SI). B) Density iso-surface (2250 kg/m3) of low density root under Ngauruhoe from unconstrained inversion (red) and geologically constrained inversion (blue). In both sections topography is shown as dark grey surface and the gravity derived basement as light grey surface. No vertical exaggeration.